[PATCH 06/10] ARM: ftrace: add Thumb-2 support
Rabin Vincent
Thumb-2 instruction set.
Cc: Catalin Marinas <catalin...@arm.com>
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/kernel/entry-common.S | 8 +++++---
1 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index 42eb166..cd2a574 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -156,7 +156,8 @@ ENTRY(__gnu_mcount_nc)
stmdb sp!, {r0-r3, lr}
ldr r0, =ftrace_trace_function
ldr r2, [r0]
- adr r0, ftrace_stub
+ THUMB( orr r2, r2, #1 )
+ adr r0, BSYM(ftrace_stub)
cmp r0, r2
bne gnu_trace
ldmia sp!, {r0-r3, ip, lr}
@@ -166,8 +167,9 @@ gnu_trace:
ldr r1, [sp, #20] @ lr of instrumented routine
mov r0, lr
sub r0, r0, #MCOUNT_INSN_SIZE
- mov lr, pc
- mov pc, r2
+ ARM( mov lr, pc )
+ ARM( mov pc, r2 )
+ THUMB( blx r2 )
ldmia sp!, {r0-r3, ip, lr}
mov pc, ip
ENDPROC(__gnu_mcount_nc)
--
1.6.6
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majo...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Rabin Vincent
ENDPROC seems to be required in order for the assembly routines to be
recognized as Thumb-2 code. If the ENDPROC annotations are not present,
calls to these routines are generated as BLX instead of BL.
Cc: Catalin Marinas <catalin...@arm.com>
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/kernel/entry-common.S | 7 ++++++-
1 files changed, 6 insertions(+), 1 deletions(-)
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index c98e3a3..42eb166 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -135,6 +135,7 @@ mcount_call:
bl ftrace_stub
ldr lr, [fp, #-4] @ restore lr
ldmia sp!, {r0-r3, pc}
+ENDPROC(mcount)
ENTRY(ftrace_caller)
stmdb sp!, {r0-r3, lr}
@@ -147,6 +148,7 @@ ftrace_call:
bl ftrace_stub
ldr lr, [fp, #-4] @ restore lr
ldmia sp!, {r0-r3, pc}
+ENDPROC(ftrace_caller)
#else
@@ -168,6 +170,7 @@ gnu_trace:
mov pc, r2
ldmia sp!, {r0-r3, ip, lr}
mov pc, ip
+ENDPROC(__gnu_mcount_nc)
#ifdef CONFIG_OLD_MCOUNT
/*
@@ -193,13 +196,15 @@ trace:
mov pc, r2
ldr lr, [fp, #-4] @ restore lr
ldmia sp!, {r0-r3, pc}
+ENDPROC(mcount)
#endif
#endif /* CONFIG_DYNAMIC_FTRACE */
.globl ftrace_stub
-ftrace_stub:
+ENTRY(ftrace_stub)
mov pc, lr
+ENDPROC(ftrace_stub)
#endif /* CONFIG_FUNCTION_TRACER */
Rabin Vincent
option. Object files built with one ABI can't be merged with object
files built with the other ABI. So, record_mcount.pl needs to use the
same compiler flags as the kernel when generating the object file with
the mcount locations. Ensure this by passing CFLAGS to the script.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
scripts/Makefile.build | 3 ++-
1 files changed, 2 insertions(+), 1 deletions(-)
diff --git a/scripts/Makefile.build b/scripts/Makefile.build
index 0b94d2f..2535c11 100644
--- a/scripts/Makefile.build
+++ b/scripts/Makefile.build
@@ -209,7 +209,8 @@ ifdef CONFIG_FTRACE_MCOUNT_RECORD
cmd_record_mcount = set -e ; perl $(srctree)/scripts/recordmcount.pl "$(ARCH)" \
"$(if $(CONFIG_CPU_BIG_ENDIAN),big,little)" \
"$(if $(CONFIG_64BIT),64,32)" \
- "$(OBJDUMP)" "$(OBJCOPY)" "$(CC)" "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
+ "$(OBJDUMP)" "$(OBJCOPY)" "$(CC) $(KBUILD_CFLAGS)" \
+ "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
"$(if $(part-of-module),1,0)" "$(@)";
endif
Rabin Vincent
with the new ftrace dyamic tracing implementation. It also adds support
for the mcount format used by newer ARM compilers.
With dynamic tracing, mcount() is implemented as a nop. Callsites are
patched on startup with nops, and dynamically patched to call to the
ftrace_caller() routine as needed.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/include/asm/ftrace.h | 19 +++++-
arch/arm/kernel/entry-common.S | 37 +++++++---
arch/arm/kernel/ftrace.c | 155 +++++++++++++++++++++++++++------------
scripts/recordmcount.pl | 2 +
4 files changed, 155 insertions(+), 58 deletions(-)
diff --git a/arch/arm/include/asm/ftrace.h b/arch/arm/include/asm/ftrace.h
index 103f7ee..4a56a2e 100644
--- a/arch/arm/include/asm/ftrace.h
+++ b/arch/arm/include/asm/ftrace.h
@@ -2,12 +2,29 @@
#define _ASM_ARM_FTRACE
#ifdef CONFIG_FUNCTION_TRACER
-#define MCOUNT_ADDR ((long)(mcount))
+#define MCOUNT_ADDR ((unsigned long)(__gnu_mcount_nc))
#define MCOUNT_INSN_SIZE 4 /* sizeof mcount call */
#ifndef __ASSEMBLY__
extern void mcount(void);
extern void __gnu_mcount_nc(void);
+
+#ifdef CONFIG_DYNAMIC_FTRACE
+struct dyn_arch_ftrace {
+#ifdef CONFIG_OLD_MCOUNT
+ bool old_mcount;
+#endif
+};
+
+static inline unsigned long ftrace_call_adjust(unsigned long addr)
+{
+ return addr;
+}
+
+extern void ftrace_caller_old(void);
+extern void ftrace_call_old(void);
+#endif
+
#endif
#endif
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index cd2a574..736cd75 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -123,32 +123,51 @@ ENDPROC(ret_from_fork)
* With both the mcount types, we need to restore the original lr before
* returning. In the __gnu_mcount_nc, version we're allowed to clobber ip.
* No other registers should be clobbered.
+ *
+ * When using dynamic ftrace, we patch out the mcount call by a "mov r0, r0"
+ * for the mcount case, and a "pop {lr}" for the __gnu_mcount_nc case (see
+ * arch/arm/kernel/ftrace.c).
*/
#ifdef CONFIG_DYNAMIC_FTRACE
-ENTRY(mcount)
+ENTRY(__gnu_mcount_nc)
+ mov ip, lr
+ ldmia sp!, {lr}
+ mov pc, ip
+ENDPROC(__gnu_mcount_nc)
+
+ENTRY(ftrace_caller)
stmdb sp!, {r0-r3, lr}
mov r0, lr
sub r0, r0, #MCOUNT_INSN_SIZE
+ ldr r1, [sp, #20]
- .globl mcount_call
-mcount_call:
+ .global ftrace_call
+ftrace_call:
bl ftrace_stub
- ldr lr, [fp, #-4] @ restore lr
- ldmia sp!, {r0-r3, pc}
+ ldmia sp!, {r0-r3, ip, lr}
+ mov pc, ip
+ENDPROC(ftrace_caller)
+
+#ifdef CONFIG_OLD_MCOUNT
+ENTRY(mcount)
+ stmdb sp!, {lr}
+ ldr lr, [fp, #-4]
+ ldmia sp!, {pc}
ENDPROC(mcount)
-ENTRY(ftrace_caller)
+ENTRY(ftrace_caller_old)
stmdb sp!, {r0-r3, lr}
ldr r1, [fp, #-4]
mov r0, lr
sub r0, r0, #MCOUNT_INSN_SIZE
- .globl ftrace_call
-ftrace_call:
+ .globl ftrace_call_old
+ftrace_call_old:
bl ftrace_stub
ldr lr, [fp, #-4] @ restore lr
ldmia sp!, {r0-r3, pc}
-ENDPROC(ftrace_caller)
+ENDPROC(ftrace_caller_old)
+#endif
#else
diff --git a/arch/arm/kernel/ftrace.c b/arch/arm/kernel/ftrace.c
index c638427..f09014c 100644
--- a/arch/arm/kernel/ftrace.c
+++ b/arch/arm/kernel/ftrace.c
@@ -2,102 +2,161 @@
* Dynamic function tracing support.
*
* Copyright (C) 2008 Abhishek Sagar <sagar.a...@gmail.com>
+ * Copyright (C) 2010 Rabin Vincent <ra...@rab.in>
*
* For licencing details, see COPYING.
*
* Defines low-level handling of mcount calls when the kernel
* is compiled with the -pg flag. When using dynamic ftrace, the
- * mcount call-sites get patched lazily with NOP till they are
- * enabled. All code mutation routines here take effect atomically.
+ * mcount call-sites get patched with NOP till they are enabled.
+ * All code mutation routines here are called under stop_machine().
*/
#include <linux/ftrace.h>
+#include <linux/uaccess.h>
#include <asm/cacheflush.h>
#include <asm/ftrace.h>
-#define PC_OFFSET 8
-#define BL_OPCODE 0xeb000000
-#define BL_OFFSET_MASK 0x00ffffff
+#define NOP 0xe8bd4000 /* pop {lr} */
-static unsigned long bl_insn;
-static const unsigned long NOP = 0xe1a00000; /* mov r0, r0 */
+#ifdef CONFIG_OLD_MCOUNT
+#define OLD_MCOUNT_ADDR ((unsigned long) mcount)
+#define OLD_FTRACE_ADDR ((unsigned long) ftrace_caller_old)
-unsigned char *ftrace_nop_replace(void)
+#define OLD_NOP 0xe1a00000 /* mov r0, r0 */
+
+static unsigned long ftrace_nop_replace(struct dyn_ftrace *rec)
+{
+ return rec->arch.old_mcount ? OLD_NOP : NOP;
+}
+
+static unsigned long adjust_address(struct dyn_ftrace *rec, unsigned long addr)
+{
+ if (!rec->arch.old_mcount)
+ return addr;
+
+ if (addr == MCOUNT_ADDR)
+ addr = OLD_MCOUNT_ADDR;
+ else if (addr == FTRACE_ADDR)
+ addr = OLD_FTRACE_ADDR;
+
+ return addr;
+}
+#else
+static unsigned long ftrace_nop_replace(struct dyn_ftrace *rec)
+{
+ return NOP;
+}
+
+static unsigned long adjust_address(struct dyn_ftrace *rec, unsigned long addr)
{
- return (char *)&NOP;
+ return addr;
}
+#endif
/* construct a branch (BL) instruction to addr */
-unsigned char *ftrace_call_replace(unsigned long pc, unsigned long addr)
+static unsigned long ftrace_call_replace(unsigned long pc, unsigned long addr)
{
long offset;
- offset = (long)addr - (long)(pc + PC_OFFSET);
+ offset = (long)addr - (long)(pc + 8);
if (unlikely(offset < -33554432 || offset > 33554428)) {
/* Can't generate branches that far (from ARM ARM). Ftrace
* doesn't generate branches outside of kernel text.
*/
WARN_ON_ONCE(1);
- return NULL;
+ return 0;
}
- offset = (offset >> 2) & BL_OFFSET_MASK;
- bl_insn = BL_OPCODE | offset;
- return (unsigned char *)&bl_insn;
-}
-int ftrace_modify_code(unsigned long pc, unsigned char *old_code,
- unsigned char *new_code)
-{
- unsigned long err = 0, replaced = 0, old, new;
+ offset = (offset >> 2) & 0x00ffffff;
- old = *(unsigned long *)old_code;
- new = *(unsigned long *)new_code;
+ return 0xeb000000 | offset;
+}
- __asm__ __volatile__ (
- "1: ldr %1, [%2] \n"
- " cmp %1, %4 \n"
- "2: streq %3, [%2] \n"
- " cmpne %1, %3 \n"
- " movne %0, #2 \n"
- "3:\n"
+static int ftrace_modify_code(unsigned long pc, unsigned long old,
+ unsigned long new)
+{
+ unsigned long replaced;
- ".section .fixup, \"ax\"\n"
- "4: mov %0, #1 \n"
- " b 3b \n"
- ".previous\n"
+ if (probe_kernel_read(&replaced, (void *)pc, MCOUNT_INSN_SIZE))
+ return -EFAULT;
- ".section __ex_table, \"a\"\n"
- " .long 1b, 4b \n"
- " .long 2b, 4b \n"
- ".previous\n"
+ if (replaced != old)
+ return -EINVAL;
- : "=r"(err), "=r"(replaced)
- : "r"(pc), "r"(new), "r"(old), "0"(err), "1"(replaced)
- : "memory");
+ if (probe_kernel_write((void *)pc, &new, MCOUNT_INSN_SIZE))
+ return -EPERM;
- if (!err && (replaced == old))
- flush_icache_range(pc, pc + MCOUNT_INSN_SIZE);
+ flush_icache_range(pc, pc + MCOUNT_INSN_SIZE);
- return err;
+ return 0;
}
int ftrace_update_ftrace_func(ftrace_func_t func)
{
- int ret;
unsigned long pc, old;
- unsigned char *new;
+ unsigned long new;
+ int ret;
pc = (unsigned long)&ftrace_call;
memcpy(&old, &ftrace_call, MCOUNT_INSN_SIZE);
new = ftrace_call_replace(pc, (unsigned long)func);
- ret = ftrace_modify_code(pc, (unsigned char *)&old, new);
+
+ ret = ftrace_modify_code(pc, old, new);
+
+#ifdef CONFIG_OLD_MCOUNT
+ if (!ret) {
+ pc = (unsigned long)&ftrace_call_old;
+ memcpy(&old, &ftrace_call_old, MCOUNT_INSN_SIZE);
+ new = ftrace_call_replace(pc, (unsigned long)func);
+
+ ret = ftrace_modify_code(pc, old, new);
+ }
+#endif
+
+ return ret;
+}
+
+int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
+{
+ unsigned long new, old;
+ unsigned long ip = rec->ip;
+
+ old = ftrace_nop_replace(rec);
+ new = ftrace_call_replace(ip, adjust_address(rec, addr));
+
+ return ftrace_modify_code(rec->ip, old, new);
+}
+
+int ftrace_make_nop(struct module *mod,
+ struct dyn_ftrace *rec, unsigned long addr)
+{
+ unsigned long ip = rec->ip;
+ unsigned long old;
+ unsigned long new;
+ int ret;
+
+ old = ftrace_call_replace(ip, adjust_address(rec, addr));
+ new = ftrace_nop_replace(rec);
+ ret = ftrace_modify_code(ip, old, new);
+
+#ifdef CONFIG_OLD_MCOUNT
+ if (ret == -EINVAL && addr == MCOUNT_ADDR) {
+ rec->arch.old_mcount = true;
+
+ old = ftrace_call_replace(ip, adjust_address(rec, addr));
+ new = ftrace_nop_replace(rec);
+ ret = ftrace_modify_code(ip, old, new);
+ }
+#endif
+
return ret;
}
-/* run from ftrace_init with irqs disabled */
int __init ftrace_dyn_arch_init(void *data)
{
- ftrace_mcount_set(data);
+ *(unsigned long *)data = 0;
+
return 0;
}
diff --git a/scripts/recordmcount.pl b/scripts/recordmcount.pl
index ea6f6e3..66c1c4b 100755
--- a/scripts/recordmcount.pl
+++ b/scripts/recordmcount.pl
@@ -267,6 +267,8 @@ if ($arch eq "x86_64") {
} elsif ($arch eq "arm") {
$alignment = 2;
$section_type = '%progbits';
+ $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_ARM_(CALL|PC24)" .
+ "\\s+(__gnu_mcount_nc|mcount)\$";
} elsif ($arch eq "ia64") {
$mcount_regex = "^\\s*([0-9a-fA-F]+):.*\\s_mcount\$";
Rabin Vincent
not allowed. However, -pg can be used to build without actually
specying -fno-omit-frame-pointer, upon which the defaut behaviour for
the target will be used.
On ARM, it is not possible to build a Thumb-2 kernel with
-fno-omit-frame-pointer (FRAME_POINTERS depends on !THUMB2_KERNEL). In
order to support ftrace for Thumb-2, we need to be able to allow a
combination of FUNCTION_TRACER and !FRAME_POINTER. We do this by
omitting -fomit-frame-pointer if ftrace is enabled.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
Makefile | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/Makefile b/Makefile
index 67d6cff..72d90de 100644
--- a/Makefile
+++ b/Makefile
@@ -546,8 +546,11 @@ endif
ifdef CONFIG_FRAME_POINTER
KBUILD_CFLAGS += -fno-omit-frame-pointer -fno-optimize-sibling-calls
else
+ifndef CONFIG_FUNCTION_TRACER
+# -fomit-frame-pointer is incompatible with -pg
KBUILD_CFLAGS += -fomit-frame-pointer
endif
+endif
ifdef CONFIG_DEBUG_INFO
KBUILD_CFLAGS += -g
Rabin Vincent
need for frame pointers. This is essential for Thumb-2 support, since
frame pointers aren't available then.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/Kconfig.debug | 5 +++++
arch/arm/kernel/armksyms.c | 2 ++
arch/arm/kernel/entry-common.S | 7 +++++++
kernel/trace/Kconfig | 2 +-
4 files changed, 15 insertions(+), 1 deletions(-)
diff --git a/arch/arm/Kconfig.debug b/arch/arm/Kconfig.debug
index 5cb9326..5501253 100644
--- a/arch/arm/Kconfig.debug
+++ b/arch/arm/Kconfig.debug
@@ -27,6 +27,11 @@ config ARM_UNWIND
the performance is not affected. Currently, this feature
only works with EABI compilers. If unsure say Y.
+config OLD_MCOUNT
+ bool
+ depends on FUNCTION_TRACER && FRAME_POINTER
+ default y
+
config DEBUG_USER
bool "Verbose user fault messages"
help
diff --git a/arch/arm/kernel/armksyms.c b/arch/arm/kernel/armksyms.c
index 8214bfe..e5e1e53 100644
--- a/arch/arm/kernel/armksyms.c
+++ b/arch/arm/kernel/armksyms.c
@@ -165,6 +165,8 @@ EXPORT_SYMBOL(_find_next_bit_be);
#endif
#ifdef CONFIG_FUNCTION_TRACER
+#ifdef CONFIG_OLD_MCOUNT
EXPORT_SYMBOL(mcount);
+#endif
EXPORT_SYMBOL(__gnu_mcount_nc);
#endif
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index d412d7c..c98e3a3 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -169,6 +169,12 @@ gnu_trace:
ldmia sp!, {r0-r3, ip, lr}
mov pc, ip
+#ifdef CONFIG_OLD_MCOUNT
+/*
+ * This is under an ifdef in order to force link-time errors for people trying
+ * to build with !FRAME_POINTER with a GCC which doesn't use the new-style
+ * mcount.
+ */
ENTRY(mcount)
stmdb sp!, {r0-r3, lr}
ldr r0, =ftrace_trace_function
@@ -187,6 +193,7 @@ trace:
mov pc, r2
ldr lr, [fp, #-4] @ restore lr
ldmia sp!, {r0-r3, pc}
+#endif
#endif /* CONFIG_DYNAMIC_FTRACE */
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 6c22d8a..7468ffe 100644
--- a/kernel/trace/Kconfig
+++ b/kernel/trace/Kconfig
@@ -126,7 +126,7 @@ if FTRACE
config FUNCTION_TRACER
bool "Kernel Function Tracer"
depends on HAVE_FUNCTION_TRACER
- select FRAME_POINTER
+ select FRAME_POINTER if (!ARM_UNWIND)
select KALLSYMS
select GENERIC_TRACER
select CONTEXT_SWITCH_TRACER
Rabin Vincent
disable dynamic ftrace for all archs that use daemon"). Now that the
code has been updated, re-enable it.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/Kconfig | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 4c33ca8..eee22bb 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -17,6 +17,8 @@ config ARM
select HAVE_KPROBES if (!XIP_KERNEL)
select HAVE_KRETPROBES if (HAVE_KPROBES)
select HAVE_FUNCTION_TRACER if (!XIP_KERNEL)
+ select HAVE_FTRACE_MCOUNT_RECORD if (!XIP_KERNEL)
+ select HAVE_DYNAMIC_FTRACE if (!XIP_KERNEL)
select HAVE_GENERIC_DMA_COHERENT
select HAVE_KERNEL_GZIP
select HAVE_KERNEL_LZO
Rabin Vincent
need to adjust the recorded mcount_loc addresses because they have the
lsb set.
Cc: Catalin Marinas <catalin...@arm.com>
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/include/asm/ftrace.h | 4 ++++
arch/arm/kernel/ftrace.c | 33 +++++++++++++++++++++++++++++++++
scripts/recordmcount.pl | 2 +-
3 files changed, 38 insertions(+), 1 deletions(-)
diff --git a/arch/arm/include/asm/ftrace.h b/arch/arm/include/asm/ftrace.h
index 4a56a2e..0845f2d 100644
--- a/arch/arm/include/asm/ftrace.h
+++ b/arch/arm/include/asm/ftrace.h
@@ -18,7 +18,11 @@ struct dyn_arch_ftrace {
static inline unsigned long ftrace_call_adjust(unsigned long addr)
{
+#ifdef CONFIG_THUMB2_KERNEL
+ return addr & ~1;
+#else
return addr;
+#endif
}
extern void ftrace_caller_old(void);
diff --git a/arch/arm/kernel/ftrace.c b/arch/arm/kernel/ftrace.c
index f09014c..971ac8c 100644
--- a/arch/arm/kernel/ftrace.c
+++ b/arch/arm/kernel/ftrace.c
@@ -18,7 +18,11 @@
#include <asm/cacheflush.h>
#include <asm/ftrace.h>
+#ifdef CONFIG_THUMB2_KERNEL
+#define NOP 0xeb04f85d /* pop.w {lr} */
+#else
#define NOP 0xe8bd4000 /* pop {lr} */
+#endif
#ifdef CONFIG_OLD_MCOUNT
#define OLD_MCOUNT_ADDR ((unsigned long) mcount)
@@ -56,6 +60,34 @@ static unsigned long adjust_address(struct dyn_ftrace *rec, unsigned long addr)
#endif
/* construct a branch (BL) instruction to addr */
+#ifdef CONFIG_THUMB2_KERNEL
+static unsigned long ftrace_call_replace(unsigned long pc, unsigned long addr)
+{
+ unsigned long s, j1, j2, i1, i2, imm10, imm11;
+ unsigned long first, second;
+ long offset;
+
+ offset = (long)addr - (long)(pc + 4);
+ if (offset < -16777216 || offset > 16777214) {
+ WARN_ON_ONCE(1);
+ return 0;
+ }
+
+ s = (offset >> 24) & 0x1;
+ i1 = (offset >> 23) & 0x1;
+ i2 = (offset >> 22) & 0x1;
+ imm10 = (offset >> 12) & 0x3ff;
+ imm11 = (offset >> 1) & 0x7ff;
+
+ j1 = (!i1) ^ s;
+ j2 = (!i2) ^ s;
+
+ first = 0xf000 | (s << 10) | imm10;
+ second = 0xd000 | (j1 << 13) | (j2 << 11) | imm11;
+
+ return (second << 16) | first;
+}
+#else
static unsigned long ftrace_call_replace(unsigned long pc, unsigned long addr)
{
long offset;
@@ -73,6 +105,7 @@ static unsigned long ftrace_call_replace(unsigned long pc, unsigned long addr)
return 0xeb000000 | offset;
}
+#endif
static int ftrace_modify_code(unsigned long pc, unsigned long old,
unsigned long new)
diff --git a/scripts/recordmcount.pl b/scripts/recordmcount.pl
index 66c1c4b..88b083b 100755
--- a/scripts/recordmcount.pl
+++ b/scripts/recordmcount.pl
@@ -267,7 +267,7 @@ if ($arch eq "x86_64") {
} elsif ($arch eq "arm") {
$alignment = 2;
$section_type = '%progbits';
- $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_ARM_(CALL|PC24)" .
+ $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_ARM_(CALL|PC24|THM_CALL)" .
"\\s+(__gnu_mcount_nc|mcount)\$";
} elsif ($arch eq "ia64") {
Rabin Vincent
from the rest of the file (and the rest of the ARM assembly in the
kernel). Clean it up.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/kernel/entry-common.S | 88 ++++++++++++++++++++--------------------
1 files changed, 44 insertions(+), 44 deletions(-)
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index 2c1db77..0b042bd 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -94,73 +94,73 @@ ENDPROC(ret_from_fork)
#ifdef CONFIG_FUNCTION_TRACER
#ifdef CONFIG_DYNAMIC_FTRACE
ENTRY(mcount)
- stmdb sp!, {r0-r3, lr}
- mov r0, lr
- sub r0, r0, #MCOUNT_INSN_SIZE
+ stmdb sp!, {r0-r3, lr}
+ mov r0, lr
+ sub r0, r0, #MCOUNT_INSN_SIZE
.globl mcount_call
mcount_call:
- bl ftrace_stub
- ldr lr, [fp, #-4] @ restore lr
- ldmia sp!, {r0-r3, pc}
+ bl ftrace_stub
+ ldr lr, [fp, #-4] @ restore lr
+ ldmia sp!, {r0-r3, pc}
ENTRY(ftrace_caller)
- stmdb sp!, {r0-r3, lr}
- ldr r1, [fp, #-4]
- mov r0, lr
- sub r0, r0, #MCOUNT_INSN_SIZE
+ stmdb sp!, {r0-r3, lr}
+ ldr r1, [fp, #-4]
+ mov r0, lr
+ sub r0, r0, #MCOUNT_INSN_SIZE
.globl ftrace_call
ftrace_call:
- bl ftrace_stub
- ldr lr, [fp, #-4] @ restore lr
- ldmia sp!, {r0-r3, pc}
+ bl ftrace_stub
+ ldr lr, [fp, #-4] @ restore lr
+ ldmia sp!, {r0-r3, pc}
#else
ENTRY(__gnu_mcount_nc)
- stmdb sp!, {r0-r3, lr}
- ldr r0, =ftrace_trace_function
- ldr r2, [r0]
- adr r0, ftrace_stub
- cmp r0, r2
- bne gnu_trace
- ldmia sp!, {r0-r3, ip, lr}
- mov pc, ip
+ stmdb sp!, {r0-r3, lr}
+ ldr r0, =ftrace_trace_function
+ ldr r2, [r0]
+ adr r0, ftrace_stub
+ cmp r0, r2
+ bne gnu_trace
+ ldmia sp!, {r0-r3, ip, lr}
+ mov pc, ip
gnu_trace:
- ldr r1, [sp, #20] @ lr of instrumented routine
- mov r0, lr
- sub r0, r0, #MCOUNT_INSN_SIZE
- mov lr, pc
- mov pc, r2
- ldmia sp!, {r0-r3, ip, lr}
- mov pc, ip
+ ldr r1, [sp, #20] @ lr of instrumented routine
+ mov r0, lr
+ sub r0, r0, #MCOUNT_INSN_SIZE
+ mov lr, pc
+ mov pc, r2
+ ldmia sp!, {r0-r3, ip, lr}
+ mov pc, ip
ENTRY(mcount)
- stmdb sp!, {r0-r3, lr}
- ldr r0, =ftrace_trace_function
- ldr r2, [r0]
- adr r0, ftrace_stub
- cmp r0, r2
- bne trace
- ldr lr, [fp, #-4] @ restore lr
- ldmia sp!, {r0-r3, pc}
+ stmdb sp!, {r0-r3, lr}
+ ldr r0, =ftrace_trace_function
+ ldr r2, [r0]
+ adr r0, ftrace_stub
+ cmp r0, r2
+ bne trace
+ ldr lr, [fp, #-4] @ restore lr
+ ldmia sp!, {r0-r3, pc}
trace:
- ldr r1, [fp, #-4] @ lr of instrumented routine
- mov r0, lr
- sub r0, r0, #MCOUNT_INSN_SIZE
- mov lr, pc
- mov pc, r2
- ldr lr, [fp, #-4] @ restore lr
- ldmia sp!, {r0-r3, pc}
+ ldr r1, [fp, #-4] @ lr of instrumented routine
+ mov r0, lr
+ sub r0, r0, #MCOUNT_INSN_SIZE
+ mov lr, pc
+ mov pc, r2
+ ldr lr, [fp, #-4] @ restore lr
+ ldmia sp!, {r0-r3, pc}
#endif /* CONFIG_DYNAMIC_FTRACE */
.globl ftrace_stub
ftrace_stub:
- mov pc, lr
+ mov pc, lr
#endif /* CONFIG_FUNCTION_TRACER */
Rabin Vincent
like.
Signed-off-by: Rabin Vincent <ra...@rab.in>
---
arch/arm/kernel/entry-common.S | 32 ++++++++++++++++++++++++++++++++
1 files changed, 32 insertions(+), 0 deletions(-)
diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
index 0b042bd..d412d7c 100644
--- a/arch/arm/kernel/entry-common.S
+++ b/arch/arm/kernel/entry-common.S
@@ -92,6 +92,38 @@ ENDPROC(ret_from_fork)
#define CALL(x) .long x
#ifdef CONFIG_FUNCTION_TRACER
+/*
+ * When compiling with -pg, gcc inserts a call to the mcount routine at the
+ * start of every function. In mcount, apart from the function's address (in
+ * lr), we need to get hold of the function's caller's address.
+ *
+ * Older GCCs (pre-4.4) inserted a call to a routine called mcount like this:
+ *
+ * bl mcount
+ *
+ * These versions have the limitation that in order for the mcount routine to
+ * be able to determine the function's caller's address, an APCS-style frame
+ * pointer (which is set up with something like the code below) is required.
+ *
+ * mov ip, sp
+ * push {fp, ip, lr, pc}
+ * sub fp, ip, #4
+ *
+ * With EABI, these frame pointers are not available unless -mapcs-frame is
+ * specified, and if building as Thumb-2, not even then.
+ *
+ * Newer GCCs (4.4+) solve this problem by introducing a new version of mcount,
+ * with call sites like:
+ *
+ * push {lr}
+ * bl __gnu_mcount_nc
+ *
+ * With these compilers, frame pointers are not necessary.
+ *
+ * With both the mcount types, we need to restore the original lr before
+ * returning. In the __gnu_mcount_nc, version we're allowed to clobber ip.
+ * No other registers should be clobbered.
+ */
#ifdef CONFIG_DYNAMIC_FTRACE
ENTRY(mcount)
stmdb sp!, {r0-r3, lr}
Uwe Kleine-König
> The mcount implementation currently uses a different indentation style
> from the rest of the file (and the rest of the ARM assembly in the
> kernel). Clean it up.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
--
Pengutronix e.K. | Uwe Kleine-K�nig |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Uwe Kleine-König
Very nice.
Maybe make the last two sentences:
In the __gnu_mcount_nc case the ip register is clobbered which is OK as
the calling convention for ARM allow clobbering this value for
subroutines and it doesn't contain parameters.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-K�nig |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Catalin Marinas
> When building as Thumb-2, the ".type foo, %function" annotation in
> ENDPROC seems to be required in order for the assembly routines to be
> recognized as Thumb-2 code. If the ENDPROC annotations are not present,
> calls to these routines are generated as BLX instead of BL.
>
> Cc: Catalin Marinas <catalin...@arm.com>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
Acked-by: Catalin Marinas <catalin...@arm.com>
--
Catalin
Catalin Marinas
> Fix the mcount routines to build and run on a kernel built with the
> Thumb-2 instruction set.
>
> Cc: Catalin Marinas <catalin...@arm.com>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
> arch/arm/kernel/entry-common.S | 8 +++++---
> 1 files changed, 5 insertions(+), 3 deletions(-)
>
> diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
> index 42eb166..cd2a574 100644
> --- a/arch/arm/kernel/entry-common.S
> +++ b/arch/arm/kernel/entry-common.S
> @@ -156,7 +156,8 @@ ENTRY(__gnu_mcount_nc)
> stmdb sp!, {r0-r3, lr}
> ldr r0, =ftrace_trace_function
> ldr r2, [r0]
> - adr r0, ftrace_stub
> + THUMB( orr r2, r2, #1 )
> + adr r0, BSYM(ftrace_stub)
> cmp r0, r2
Does this code not give the correct result if not modified?
> bne gnu_trace
> ldmia sp!, {r0-r3, ip, lr}
> @@ -166,8 +167,9 @@ gnu_trace:
> ldr r1, [sp, #20] @ lr of instrumented routine
> mov r0, lr
> sub r0, r0, #MCOUNT_INSN_SIZE
> - mov lr, pc
> - mov pc, r2
> + ARM( mov lr, pc )
> + ARM( mov pc, r2 )
> + THUMB( blx r2 )
> ldmia sp!, {r0-r3, ip, lr}
> mov pc, ip
> ENDPROC(__gnu_mcount_nc)
As above, what does this need modifying? "mov pc, r2" wouldn't change
the mode to ARM even if the value in r2 is even. It may need THUMB(nop)
after this instruction.
I could of course miss something as I haven't actually tried this code.
--
Catalin
Uwe Kleine-König
On Sun, Feb 14, 2010 at 01:18:36AM +0530, Rabin Vincent wrote:
> This adds mcount recording and updates dynamic ftrace for ARM to work
> with the new ftrace dyamic tracing implementation. It also adds support
> for the mcount format used by newer ARM compilers.
>
> With dynamic tracing, mcount() is implemented as a nop. Callsites are
> patched on startup with nops, and dynamically patched to call to the
> ftrace_caller() routine as needed.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
> arch/arm/include/asm/ftrace.h | 19 +++++-
> arch/arm/kernel/entry-common.S | 37 +++++++---
> arch/arm/kernel/ftrace.c | 155 +++++++++++++++++++++++++++------------
> scripts/recordmcount.pl | 2 +
> 4 files changed, 155 insertions(+), 58 deletions(-)
>
> diff --git a/arch/arm/include/asm/ftrace.h b/arch/arm/include/asm/ftrace.h
> index 103f7ee..4a56a2e 100644
> --- a/arch/arm/include/asm/ftrace.h
> +++ b/arch/arm/include/asm/ftrace.h
> @@ -2,12 +2,29 @@
> #define _ASM_ARM_FTRACE
>
> #ifdef CONFIG_FUNCTION_TRACER
> -#define MCOUNT_ADDR ((long)(mcount))
> +#define MCOUNT_ADDR ((unsigned long)(__gnu_mcount_nc))
hhmm, does this work with both the old and the new mcount function?
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-K�nig |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Rabin Vincent
> > diff --git a/arch/arm/include/asm/ftrace.h b/arch/arm/include/asm/ftrace.h
> > index 103f7ee..4a56a2e 100644
> > --- a/arch/arm/include/asm/ftrace.h
> > +++ b/arch/arm/include/asm/ftrace.h
> > @@ -2,12 +2,29 @@
> > #define _ASM_ARM_FTRACE
> >
> > #ifdef CONFIG_FUNCTION_TRACER
> > -#define MCOUNT_ADDR ((long)(mcount))
> > +#define MCOUNT_ADDR ((unsigned long)(__gnu_mcount_nc))
> hhmm, does this work with both the old and the new mcount function?
Yes, MCOUNT_ADDR is not used by the generic ftrace code directly, but is
instead only passed to our arch specific functions. We determine the
right mcount address to use there:
> +#define OLD_MCOUNT_ADDR ((unsigned long) mcount)
[..]
> +static unsigned long adjust_address(struct dyn_ftrace *rec, unsigned long addr)
> +{
> + if (!rec->arch.old_mcount)
> + return addr;
> +
> + if (addr == MCOUNT_ADDR)
> + addr = OLD_MCOUNT_ADDR;
> + else if (addr == FTRACE_ADDR)
> + addr = OLD_FTRACE_ADDR;
> +
> + return addr;
> +}
I changed MCOUNT_ADDR from mcount to __gnu_mcount_nc because we always
build __gnu_mcount_nc in, but don't build mcount when using Thumb-2.
Rabin
Rabin Vincent
> >diff --git a/arch/arm/kernel/entry-common.S b/arch/arm/kernel/entry-common.S
> >index 42eb166..cd2a574 100644
> >--- a/arch/arm/kernel/entry-common.S
> >+++ b/arch/arm/kernel/entry-common.S
> >@@ -156,7 +156,8 @@ ENTRY(__gnu_mcount_nc)
> > stmdb sp!, {r0-r3, lr}
> > ldr r0, =ftrace_trace_function
> > ldr r2, [r0]
> >- adr r0, ftrace_stub
> >+ THUMB( orr r2, r2, #1 )
> >+ adr r0, BSYM(ftrace_stub)
> > cmp r0, r2
>
> Does this code not give the correct result if not modified?
Without the BSYM, I get assembler errors:
entry-common.S: Assembler messages:
entry-common.S:179: Error: invalid immediate for address calculation (value = 0x00000004)
Without the orr, the lsb is not set on the pointer loaded from
ftrace_trace_function, but is set on BSYM(ftrace_stub), leading to the
comparison failing even when the pointer is pointing to ftrace_stub.
>
> > bne gnu_trace
> > ldmia sp!, {r0-r3, ip, lr}
> >@@ -166,8 +167,9 @@ gnu_trace:
> > ldr r1, [sp, #20] @ lr of instrumented routine
> > mov r0, lr
> > sub r0, r0, #MCOUNT_INSN_SIZE
> >- mov lr, pc
> >- mov pc, r2
> >+ ARM( mov lr, pc )
> >+ ARM( mov pc, r2 )
> >+ THUMB( blx r2 )
> > ldmia sp!, {r0-r3, ip, lr}
> > mov pc, ip
> > ENDPROC(__gnu_mcount_nc)
>
> As above, what does this need modifying? "mov pc, r2" wouldn't
> change the mode to ARM even if the value in r2 is even. It may need
> THUMB(nop) after this instruction.
The "mov pc, r2" is not the problem. The problem is the "mov lr, pc",
which does not set the lsb when storing the pc in lr. The called
function returns with "bx lr", and the mode changes to ARM. The blx is
to avoid this.
Rabin
Rabin Vincent
> On Sun, Feb 14, 2010 at 01:18:30AM +0530, Rabin Vincent wrote:
> > + * With both the mcount types, we need to restore the original lr before
> > + * returning. In the __gnu_mcount_nc, version we're allowed to clobber ip.
> > + * No other registers should be clobbered.
> > + */
> Very nice.
>
> Maybe make the last two sentences:
>
> In the __gnu_mcount_nc case the ip register is clobbered which is OK as
> the calling convention for ARM allow clobbering this value for
> subroutines and it doesn't contain parameters.
Won't quoting calling conventions here be misleading? The mcounts don't
following normal convention for the other bits (we can't clobber even
the registers that are normally caller-saved, we need to restore lr,
etc.)
Rabin
Frederic Weisbecker
> from the rest of the file (and the rest of the ARM assembly in the
> kernel). Clean it up.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
Acked-by: Frederic Weisbecker <fwei...@gmail.com>
Frederic Weisbecker
> With current gcc, compiling with both -pg and -fomit-frame-pointer is
> not allowed. However, -pg can be used to build without actually
> specying -fno-omit-frame-pointer, upon which the defaut behaviour for
> the target will be used.
Doh!....
> On ARM, it is not possible to build a Thumb-2 kernel with
> -fno-omit-frame-pointer (FRAME_POINTERS depends on !THUMB2_KERNEL). In
> order to support ftrace for Thumb-2, we need to be able to allow a
> combination of FUNCTION_TRACER and !FRAME_POINTER. We do this by
> omitting -fomit-frame-pointer if ftrace is enabled.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
I might be missing something subtle on this one, but it looks
good to me.
Acked-by: Frederic Weisbecker <fwei...@gmail.com>
Frederic Weisbecker
> like.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
Cool, this was really missing I think.
Acked-by: Frederic Weisbecker <fwei...@gmail.com>
Frederic Weisbecker
> With a new enough GCC, ARM function tracing can be supported without the
> need for frame pointers. This is essential for Thumb-2 support, since
> frame pointers aren't available then.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
> arch/arm/Kconfig.debug | 5 +++++
> arch/arm/kernel/armksyms.c | 2 ++
> arch/arm/kernel/entry-common.S | 7 +++++++
> kernel/trace/Kconfig | 2 +-
> 4 files changed, 15 insertions(+), 1 deletions(-)
>
> diff --git a/arch/arm/Kconfig.debug b/arch/arm/Kconfig.debug
> index 5cb9326..5501253 100644
> --- a/arch/arm/Kconfig.debug
> +++ b/arch/arm/Kconfig.debug
> @@ -27,6 +27,11 @@ config ARM_UNWIND
> the performance is not affected. Currently, this feature
> only works with EABI compilers. If unsure say Y.
>
> +config OLD_MCOUNT
> + bool
> + depends on FUNCTION_TRACER && FRAME_POINTER
> + default y
> +
Btw, you described that mcount is used in some gcc versions and
__gnu_mcount_nc in newers.
Shouldn't we have a gcc version dependency here? (not sure we can
do this from Kconfig though).
So, if I understand well. If people have ARM_UNWIND but
FUNCTION_TRACER, it might crash at link time in case
they don't have a recent enough gcc version to
support the new -pg style?
That doesn't look good.
Ideally, we need HAVE_FUNCTION_TRACER to be enabled in arm
only if (old gcc && frame pointers) || new gcc
And then, we need config OLD_MCOUNT:
old gcc && FUNCTION_TRACER
and config NEW_MCOUNT:
new gcc && FUNCTION_TRACER
so that we can selectively build mcount or __gnu_mcount_nc.
Hmm, I fear we can't check gcc version from Kconfig, as I'm
grepping on Kconfig files...
Uwe Kleine-König
> On Sat, Feb 13, 2010 at 09:37:48PM +0100, Uwe Kleine-K�nig wrote:
> > On Sun, Feb 14, 2010 at 01:18:30AM +0530, Rabin Vincent wrote:
> > > + * With both the mcount types, we need to restore the original lr before
> > > + * returning. In the __gnu_mcount_nc, version we're allowed to clobber ip.
> > > + * No other registers should be clobbered.
> > > + */
> > Very nice.
> >
> > Maybe make the last two sentences:
> >
> > In the __gnu_mcount_nc case the ip register is clobbered which is OK as
> > the calling convention for ARM allow clobbering this value for
> > subroutines and it doesn't contain parameters.
>
> Won't quoting calling conventions here be misleading? The mcounts don't
> following normal convention for the other bits (we can't clobber even
> the registers that are normally caller-saved, we need to restore lr,
> etc.)
must be transparent for both the caller and the callee. As the called
function needs to know the return address, mcount must not clobber it.
And as ip is don't care and can be clobbered mcount can use it as it
likes.
Best regards
Uwe
--
Pengutronix e.K. | Uwe Kleine-K�nig |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Steven Rostedt
>
> Btw, you described that mcount is used in some gcc versions and
> __gnu_mcount_nc in newers.
> Shouldn't we have a gcc version dependency here? (not sure we can
> do this from Kconfig though).
>
Maybe we can add something to recordmcount.pl?
I would not check gcc version through KCONFIG, but you can have the gcc
version checked and define another macro in a header somewhere, like
asm/ftrace.h. Then we could do something different while it is
compiling.
Disclaimer:
I have to go back and read the entire thread, to know what exactly is
going on ;-)
-- Steve
Steven Rostedt
> On ARM, we have two ABIs, and the ABI used is controlled via a config
> option. Object files built with one ABI can't be merged with object
> files built with the other ABI. So, record_mcount.pl needs to use the
> same compiler flags as the kernel when generating the object file with
> the mcount locations. Ensure this by passing CFLAGS to the script.
I'm fine with this for now, but I'm wondering if we should pass the
autoconf.h to the script and parse that instead. This would give us all
set config options that we can look at.
But I need to go back and start looking at converting recordmcount.pl to
recordmcount.c again.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
> scripts/Makefile.build | 3 ++-
> 1 files changed, 2 insertions(+), 1 deletions(-)
>
> diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> index 0b94d2f..2535c11 100644
> --- a/scripts/Makefile.build
> +++ b/scripts/Makefile.build
> @@ -209,7 +209,8 @@ ifdef CONFIG_FTRACE_MCOUNT_RECORD
> cmd_record_mcount = set -e ; perl $(srctree)/scripts/recordmcount.pl "$(ARCH)" \
> "$(if $(CONFIG_CPU_BIG_ENDIAN),big,little)" \
> "$(if $(CONFIG_64BIT),64,32)" \
> - "$(OBJDUMP)" "$(OBJCOPY)" "$(CC)" "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
> + "$(OBJDUMP)" "$(OBJCOPY)" "$(CC) $(KBUILD_CFLAGS)" \
> + "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
Doesn't this need to be with the change of recordmcount.pl too?
Otherwise applying this patch alone will break it. We need every step of
the patches to work otherwise we risk breaking a kernel bisect.
-- Steve
> "$(if $(part-of-module),1,0)" "$(@)";
> endif
>
--
Steven Rostedt
> need to adjust the recorded mcount_loc addresses because they have the
> lsb set.
>
> Cc: Catalin Marinas <catalin...@arm.com>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
> arch/arm/include/asm/ftrace.h | 4 ++++
> arch/arm/kernel/ftrace.c | 33 +++++++++++++++++++++++++++++++++
> scripts/recordmcount.pl | 2 +-
> 3 files changed, 38 insertions(+), 1 deletions(-)
>
> diff --git a/arch/arm/include/asm/ftrace.h b/arch/arm/include/asm/ftrace.h
> index 4a56a2e..0845f2d 100644
> --- a/arch/arm/include/asm/ftrace.h
> +++ b/arch/arm/include/asm/ftrace.h
> @@ -18,7 +18,11 @@ struct dyn_arch_ftrace {
>
> static inline unsigned long ftrace_call_adjust(unsigned long addr)
> {
> +#ifdef CONFIG_THUMB2_KERNEL
> + return addr & ~1;
Would it be safe to do that for all arm archs? Instead of adding a
config option around it. Or do some arm archs need the lsb set.
I'm just saying a comment here would look better than #ifdef #else.
For the recordmcount.pl change;
Acked-by: Steven Rostedt <ros...@goodmis.org>
-- Steve
> @@ -267,7 +267,7 @@ if ($arch eq "x86_64") {
> } elsif ($arch eq "arm") {
> $alignment = 2;
> $section_type = '%progbits';
> - $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_ARM_(CALL|PC24)" .
> + $mcount_regex = "^\\s*([0-9a-fA-F]+):\\s*R_ARM_(CALL|PC24|THM_CALL)" .
> "\\s+(__gnu_mcount_nc|mcount)\$";
>
> } elsif ($arch eq "ia64") {
--
Steven Rostedt
> need for frame pointers. This is essential for Thumb-2 support, since
> frame pointers aren't available then.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
>
> diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
> index 6c22d8a..7468ffe 100644
> --- a/kernel/trace/Kconfig
> +++ b/kernel/trace/Kconfig
> @@ -126,7 +126,7 @@ if FTRACE
> config FUNCTION_TRACER
> bool "Kernel Function Tracer"
> depends on HAVE_FUNCTION_TRACER
> - select FRAME_POINTER
> + select FRAME_POINTER if (!ARM_UNWIND)
> select KALLSYMS
> select GENERIC_TRACER
> select CONTEXT_SWITCH_TRACER
Ah, evolution is keeping these patches out of order, so I did not see
this change before commenting about the change to the Makefile.
A better comment is needed in the Makefile.
-- Steve
Steven Rostedt
> not allowed. However, -pg can be used to build without actually
> specying -fno-omit-frame-pointer, upon which the defaut behaviour for
> the target will be used.
>
> On ARM, it is not possible to build a Thumb-2 kernel with
> -fno-omit-frame-pointer (FRAME_POINTERS depends on !THUMB2_KERNEL). In
> order to support ftrace for Thumb-2, we need to be able to allow a
> combination of FUNCTION_TRACER and !FRAME_POINTER. We do this by
> omitting -fomit-frame-pointer if ftrace is enabled.
>
> Signed-off-by: Rabin Vincent <ra...@rab.in>
> ---
> Makefile | 3 +++
> 1 files changed, 3 insertions(+), 0 deletions(-)
>
> diff --git a/Makefile b/Makefile
> index 67d6cff..72d90de 100644
> --- a/Makefile
> +++ b/Makefile
> @@ -546,8 +546,11 @@ endif
> ifdef CONFIG_FRAME_POINTER
> KBUILD_CFLAGS += -fno-omit-frame-pointer -fno-optimize-sibling-calls
> else
> +ifndef CONFIG_FUNCTION_TRACER
Is this needed? Selecting FUNCTION_TRACER also selects FRAME_POINTER,
which this change is in the else part of that #if.
-- Steve
> +# -fomit-frame-pointer is incompatible with -pg
> KBUILD_CFLAGS += -fomit-frame-pointer
> endif
> +endif
>
> ifdef CONFIG_DEBUG_INFO
> KBUILD_CFLAGS += -g
--
Rabin Vincent
> On Sun, 2010-02-14 at 01:18 +0530, Rabin Vincent wrote:
> > --- a/arch/arm/include/asm/ftrace.h
> > +++ b/arch/arm/include/asm/ftrace.h
> > @@ -18,7 +18,11 @@ struct dyn_arch_ftrace {
> >
> > static inline unsigned long ftrace_call_adjust(unsigned long addr)
> > {
> > +#ifdef CONFIG_THUMB2_KERNEL
> > + return addr & ~1;
>
> Would it be safe to do that for all arm archs? Instead of adding a
> config option around it. Or do some arm archs need the lsb set.
>
> I'm just saying a comment here would look better than #ifdef #else.
You're right; I'll change this. It's safe even when we build for the
ARM instruction set.
Rabin
Frederic Weisbecker
> On Mon, 2010-02-22 at 20:05 +0100, Frederic Weisbecker wrote:
>
> >
> > Btw, you described that mcount is used in some gcc versions and
> > __gnu_mcount_nc in newers.
> > Shouldn't we have a gcc version dependency here? (not sure we can
> > do this from Kconfig though).
> >
>
> Maybe we can add something to recordmcount.pl?
May be yeah, with an explicit message that describes the issue.
Yeah. I understand now why we can't do that on Kconfig time, HOSTCC
can be different than CC.
The most important thing is to have a message that describes the issue
if the compiler-config combination is impossible.
Rabin Vincent
> On Tue, Feb 23, 2010 at 08:18:33AM -0500, Steven Rostedt wrote:
> > On Mon, 2010-02-22 at 20:05 +0100, Frederic Weisbecker wrote:
> > >
> > > Btw, you described that mcount is used in some gcc versions and
> > > __gnu_mcount_nc in newers.
> > > Shouldn't we have a gcc version dependency here? (not sure we can
> > > do this from Kconfig though).
> > >
> > Maybe we can add something to recordmcount.pl?
>
> May be yeah, with an explicit message that describes the issue.
Adding something to recordmcount.pl to detect this would be
insufficient, because this affects non-dynamic ftrace too.
They'd probably get an "undefined reference to mcount" message. I
forced the link time error because otherwise they'll have runtime
problems.
> > >
> > > That doesn't look good.
> > >
> > > Ideally, we need HAVE_FUNCTION_TRACER to be enabled in arm
> > > only if (old gcc && frame pointers) || new gcc
> > >
> > > And then, we need config OLD_MCOUNT:
> > > old gcc && FUNCTION_TRACER
> > >
> > > and config NEW_MCOUNT:
> > > new gcc && FUNCTION_TRACER
> > >
> > > so that we can selectively build mcount or __gnu_mcount_nc.
There's more to this. When we build with FRAME_POINTER, we need to
always build in both version of mcount. This is to allow modules to be
built with different versions of the compiler (one which uses the old
mcount and one with the new mcount).
> > >
> > > Hmm, I fear we can't check gcc version from Kconfig, as I'm
> > > grepping on Kconfig files...
> >
> >
> > I would not check gcc version through KCONFIG, but you can have the gcc
> > version checked and define another macro in a header somewhere, like
> > asm/ftrace.h. Then we could do something different while it is
> > compiling.
>
> Yeah. I understand now why we can't do that on Kconfig time, HOSTCC
> can be different than CC.
>
> The most important thing is to have a message that describes the issue
> if the compiler-config combination is impossible.
The easiest option to have a more obvious message than a linker error
would be to add something like this in entry-common.S:
#if (__GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 4)) \
&& !defined(OLD_MCOUNT)
#warning Ftrace on GCC < 4.4 requires frame pointers
#endif
Messages like this are also present in other places in ARM (for example
arch/arm/kernel/unwind.c), so this wouldn't be this first.
Note that the above message would help in all cases except the one where
someone builds the kernel with !fp and GCC 4.4+, and then builds a
module with an older GCC. That would still be only a linker error.
Rabin
Steven Rostedt
> The easiest option to have a more obvious message than a linker error
> would be to add something like this in entry-common.S:
>
> #if (__GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 4)) \
> && !defined(OLD_MCOUNT)
> #warning Ftrace on GCC < 4.4 requires frame pointers
> #endif
>
> Messages like this are also present in other places in ARM (for example
> arch/arm/kernel/unwind.c), so this wouldn't be this first.
>
> Note that the above message would help in all cases except the one where
> someone builds the kernel with !fp and GCC 4.4+, and then builds a
> module with an older GCC. That would still be only a linker error.
I was going to recommend the #if above. But shouldn't it be a #error
instead of a #warning?
-- Steve
Rabin Vincent
> On Sun, 2010-02-14 at 01:18 +0530, Rabin Vincent wrote:
> > On ARM, we have two ABIs, and the ABI used is controlled via a config
> > option. Object files built with one ABI can't be merged with object
> > files built with the other ABI. So, record_mcount.pl needs to use the
> > same compiler flags as the kernel when generating the object file with
> > the mcount locations. Ensure this by passing CFLAGS to the script.
>
> I'm fine with this for now, but I'm wondering if we should pass the
> autoconf.h to the script and parse that instead. This would give us all
> set config options that we can look at.
The correct ABI options are constructed using cc-option and friends, and
that logic would need to be duplicated in recordmcount.pl if we had to
parse autoconf.h.
I originally tried to add a new RECORDMCOUNT_CFLAGS variable in the ARM
Makefile and have that used here instead of KBUILD_CFLAGS (since
KBUILD_CFLAGS includes a whole bunch of possibly irrelevant options),
but coudn't figure out why my new variable wasn't accessible in
Makefile.build.
>
> But I need to go back and start looking at converting recordmcount.pl to
> recordmcount.c again.
>
> >
> > Signed-off-by: Rabin Vincent <ra...@rab.in>
> > ---
> > scripts/Makefile.build | 3 ++-
> > 1 files changed, 2 insertions(+), 1 deletions(-)
> >
> > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > index 0b94d2f..2535c11 100644
> > --- a/scripts/Makefile.build
> > +++ b/scripts/Makefile.build
> > @@ -209,7 +209,8 @@ ifdef CONFIG_FTRACE_MCOUNT_RECORD
> > cmd_record_mcount = set -e ; perl $(srctree)/scripts/recordmcount.pl "$(ARCH)" \
> > "$(if $(CONFIG_CPU_BIG_ENDIAN),big,little)" \
> > "$(if $(CONFIG_64BIT),64,32)" \
> > - "$(OBJDUMP)" "$(OBJCOPY)" "$(CC)" "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
> > + "$(OBJDUMP)" "$(OBJCOPY)" "$(CC) $(KBUILD_CFLAGS)" \
> > + "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
>
> Doesn't this need to be with the change of recordmcount.pl too?
> Otherwise applying this patch alone will break it. We need every step of
> the patches to work otherwise we risk breaking a kernel bisect.
Dynamic ftrace for ARM is only enabled in the last patch of the series,
so this woudn't be breaking it. No modification should be required for
record_mcount.pl for other archs because this doesn't add/remove any
parameters.
Rabin
Rabin Vincent
> On Tue, 2010-02-23 at 23:28 +0530, Rabin Vincent wrote:
> > The easiest option to have a more obvious message than a linker error
> > would be to add something like this in entry-common.S:
> >
> > #if (__GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 4)) \
> > && !defined(OLD_MCOUNT)
> > #warning Ftrace on GCC < 4.4 requires frame pointers
> > #endif
> >
> > Messages like this are also present in other places in ARM (for example
> > arch/arm/kernel/unwind.c), so this wouldn't be this first.
> >
> > Note that the above message would help in all cases except the one where
> > someone builds the kernel with !fp and GCC 4.4+, and then builds a
> > module with an older GCC. That would still be only a linker error.
>
> I was going to recommend the #if above. But shouldn't it be a #error
> instead of a #warning?
Yes, I'll make it a #error.
(I had it at #warning in case some one had a custom compiler that had
the new mcount patch backported, and chose to ignore the warning. But
it's probably better to leave it at #error and see if someone actually
complains about it, rather than have someone miss the message with it as
a #warning.)
Rabin
Steven Rostedt
> > > diff --git a/scripts/Makefile.build b/scripts/Makefile.build
> > > index 0b94d2f..2535c11 100644
> > > --- a/scripts/Makefile.build
> > > +++ b/scripts/Makefile.build
> > > @@ -209,7 +209,8 @@ ifdef CONFIG_FTRACE_MCOUNT_RECORD
> > > cmd_record_mcount = set -e ; perl $(srctree)/scripts/recordmcount.pl "$(ARCH)" \
> > > "$(if $(CONFIG_CPU_BIG_ENDIAN),big,little)" \
> > > "$(if $(CONFIG_64BIT),64,32)" \
> > > - "$(OBJDUMP)" "$(OBJCOPY)" "$(CC)" "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
> > > + "$(OBJDUMP)" "$(OBJCOPY)" "$(CC) $(KBUILD_CFLAGS)" \
> > > + "$(LD)" "$(NM)" "$(RM)" "$(MV)" \
> >
> > Doesn't this need to be with the change of recordmcount.pl too?
> > Otherwise applying this patch alone will break it. We need every step of
> > the patches to work otherwise we risk breaking a kernel bisect.
>
> Dynamic ftrace for ARM is only enabled in the last patch of the series,
> so this woudn't be breaking it. No modification should be required for
> record_mcount.pl for other archs because this doesn't add/remove any
> parameters.
OK, it looked to me that "KBUILD_CFLAGS" was a separate parameter. I
miscounted my quotes. ;-)
-- Steve
Rabin Vincent
> On Mon, Feb 22, 2010 at 11:36:24PM +0530, Rabin Vincent wrote:
> > On Sat, Feb 13, 2010 at 09:37:48PM +0100, Uwe Kleine-K�nig wrote:
> > > On Sun, Feb 14, 2010 at 01:18:30AM +0530, Rabin Vincent wrote:
> > > > + * With both the mcount types, we need to restore the original lr before
> > > > + * returning. In the __gnu_mcount_nc, version we're allowed to clobber ip.
> > > > + * No other registers should be clobbered.
> > > > + */
> > > Very nice.
> > >
> > > Maybe make the last two sentences:
> > >
> > > In the __gnu_mcount_nc case the ip register is clobbered which is OK as
> > > the calling convention for ARM allow clobbering this value for
> > > subroutines and it doesn't contain parameters.
> >
> > Won't quoting calling conventions here be misleading? The mcounts don't
> > following normal convention for the other bits (we can't clobber even
> > the registers that are normally caller-saved, we need to restore lr,
> > etc.)
> actually mcount is a function in the middle of a function call. So it
> must be transparent for both the caller and the callee. As the called
> function needs to know the return address, mcount must not clobber it.
> And as ip is don't care and can be clobbered mcount can use it as it
> likes.
OK. How about this for the last paragraph:
mcount can be thought of as a function called in the middle of a
subroutine call. As such, it needs to be transparent for both the
caller and the callee: the original lr needs to be restored when
leaving mcount, and no registers should be clobbered. (In the
__gnu_mcount_nc implementation, we clobber the ip register. This
is OK because the ARM calling convention allows it to be clobbered
in subroutines and doesn't use it to hold parameters.)
Rabin
Uwe Kleine-König
> OK. How about this for the last paragraph:
>
> mcount can be thought of as a function called in the middle of a
> subroutine call. As such, it needs to be transparent for both the
> caller and the callee: the original lr needs to be restored when
> leaving mcount, and no registers should be clobbered. (In the
> __gnu_mcount_nc implementation, we clobber the ip register. This
> is OK because the ARM calling convention allows it to be clobbered
> in subroutines and doesn't use it to hold parameters.)
I rechecked the document describing the calling convention. It writes
about ip:
Both the ARM- and Thumb-state BL instructions are unable to address the
full 32-bit address space, so it may be necessary for the linker to
insert a veneer between the calling routine and the called subroutine.
Veneers may also be needed to support ARM-Thumb inter-working or dynamic
linking. Any veneer inserted must preserve the contents of all
registers except IP (r12) and the condition code flags; a conforming
program must assume that a veneer that alters IP may be inserted at any
branch instruction that is exposed to a relocation that supports inter-
working or long branches.
So the last sentence isn't completly accurate. But describing the
actual situation is a bit overkill, so I'm OK with your suggestion.
Best regards and thanks
Uwe
--
Pengutronix e.K. | Uwe Kleine-K�nig |
Industrial Linux Solutions | http://www.pengutronix.de/ |