parisc: Fix ftrace function tracer

Fix the FTRACE function tracer for 32- and 64-bit kernel.
The former code was horribly broken.

Reimplement most coding in assembly and utilize optimizations, e.g. put
mcount() and ftrace_stub() into one L1 cacheline.

Signed-off-by: Helge Deller <deller@gmx.de>
This commit is contained in:
Helge Deller 2016-04-13 22:27:22 +02:00
parent b562e44f50
commit 366dd4ea9d
8 changed files with 117 additions and 171 deletions

View File

@ -4,8 +4,8 @@ config PARISC
select ARCH_MIGHT_HAVE_PC_PARPORT
select HAVE_IDE
select HAVE_OPROFILE
select HAVE_FUNCTION_TRACER if 64BIT
select HAVE_FUNCTION_GRAPH_TRACER if 64BIT
select HAVE_FUNCTION_TRACER
select HAVE_FUNCTION_GRAPH_TRACER
select ARCH_WANT_FRAME_POINTERS
select RTC_CLASS
select RTC_DRV_GENERIC

View File

@ -2,9 +2,13 @@ menu "Kernel hacking"
source "lib/Kconfig.debug"
config TRACE_IRQFLAGS_SUPPORT
def_bool y
config DEBUG_RODATA
bool "Write protect kernel read-only data structures"
depends on DEBUG_KERNEL
default y
help
Mark the kernel read-only data as write-protected in the pagetables,
in order to catch accidental (and incorrect) writes to such const

View File

@ -62,9 +62,7 @@ cflags-y += -mdisable-fpregs
# Without this, "ld -r" results in .text sections that are too big
# (> 0x40000) for branches to reach stubs.
ifndef CONFIG_FUNCTION_TRACER
cflags-y += -ffunction-sections
endif
cflags-y += -ffunction-sections
# Use long jumps instead of long branches (needed if your linker fails to
# link a too big vmlinux executable). Not enabled for building modules.

View File

@ -4,23 +4,7 @@
#ifndef __ASSEMBLY__
extern void mcount(void);
/*
* Stack of return addresses for functions of a thread.
* Used in struct thread_info
*/
struct ftrace_ret_stack {
unsigned long ret;
unsigned long func;
unsigned long long calltime;
};
/*
* Primary handler of a function return.
* It relays on ftrace_return_to_handler.
* Defined in entry.S
*/
extern void return_to_handler(void);
#define MCOUNT_INSN_SIZE 4
extern unsigned long return_address(unsigned int);

View File

@ -15,11 +15,7 @@ ifdef CONFIG_FUNCTION_TRACER
# Do not profile debug and lowlevel utilities
CFLAGS_REMOVE_ftrace.o = -pg
CFLAGS_REMOVE_cache.o = -pg
CFLAGS_REMOVE_irq.o = -pg
CFLAGS_REMOVE_pacache.o = -pg
CFLAGS_REMOVE_perf.o = -pg
CFLAGS_REMOVE_traps.o = -pg
CFLAGS_REMOVE_unaligned.o = -pg
CFLAGS_REMOVE_unwind.o = -pg
endif

View File

@ -1970,43 +1970,98 @@ pt_regs_ok:
b intr_restore
copy %r25,%r16
.import schedule,code
syscall_do_resched:
BL schedule,%r2
load32 syscall_check_resched,%r2 /* if resched, we start over again */
load32 schedule,%r19
bv %r0(%r19) /* jumps to schedule() */
#ifdef CONFIG_64BIT
ldo -16(%r30),%r29 /* Reference param save area */
#else
nop
#endif
b syscall_check_resched /* if resched, we start over again */
nop
ENDPROC(syscall_exit)
#ifdef CONFIG_FUNCTION_TRACER
.import ftrace_function_trampoline,code
ENTRY(_mcount)
copy %r3, %arg2
.align L1_CACHE_BYTES
.globl mcount
.type mcount, @function
ENTRY(mcount)
_mcount:
.export _mcount,data
.proc
.callinfo caller,frame=0
.entry
/*
* The 64bit mcount() function pointer needs 4 dwords, of which the
* first two are free. We optimize it here and put 2 instructions for
* calling mcount(), and 2 instructions for ftrace_stub(). That way we
* have all on one L1 cacheline.
*/
b ftrace_function_trampoline
nop
ENDPROC(_mcount)
ENTRY(return_to_handler)
load32 return_trampoline, %rp
copy %ret0, %arg0
copy %ret1, %arg1
b ftrace_return_to_handler
nop
return_trampoline:
copy %ret0, %rp
copy %r23, %ret0
copy %r24, %ret1
.globl ftrace_stub
copy %r3, %arg2 /* caller original %sp */
ftrace_stub:
.globl ftrace_stub
.type ftrace_stub, @function
#ifdef CONFIG_64BIT
bve (%rp)
#else
bv %r0(%rp)
#endif
nop
#ifdef CONFIG_64BIT
.dword mcount
.dword 0 /* code in head.S puts value of global gp here */
#endif
.exit
.procend
ENDPROC(mcount)
.align 8
.globl return_to_handler
.type return_to_handler, @function
ENTRY(return_to_handler)
.proc
.callinfo caller,frame=FRAME_SIZE
.entry
.export parisc_return_to_handler,data
parisc_return_to_handler:
copy %r3,%r1
STREG %r0,-RP_OFFSET(%sp) /* store 0 as %rp */
copy %sp,%r3
STREGM %r1,FRAME_SIZE(%sp)
STREG %ret0,8(%r3)
STREG %ret1,16(%r3)
#ifdef CONFIG_64BIT
loadgp
#endif
/* call ftrace_return_to_handler(0) */
#ifdef CONFIG_64BIT
ldo -16(%sp),%ret1 /* Reference param save area */
#endif
BL ftrace_return_to_handler,%r2
ldi 0,%r26
copy %ret0,%rp
/* restore original return values */
LDREG 8(%r3),%ret0
LDREG 16(%r3),%ret1
/* return from function */
#ifdef CONFIG_64BIT
bve (%rp)
#else
bv %r0(%rp)
#endif
LDREGM -FRAME_SIZE(%sp),%r3
.exit
.procend
ENDPROC(return_to_handler)
#endif /* CONFIG_FUNCTION_TRACER */
#ifdef CONFIG_IRQSTACKS

View File

@ -1,6 +1,6 @@
/*
* Code for tracing calls in Linux kernel.
* Copyright (C) 2009 Helge Deller <deller@gmx.de>
* Copyright (C) 2009-2016 Helge Deller <deller@gmx.de>
*
* based on code for x86 which is:
* Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
@ -13,104 +13,21 @@
#include <linux/init.h>
#include <linux/ftrace.h>
#include <asm/assembly.h>
#include <asm/sections.h>
#include <asm/ftrace.h>
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
/* Add a function return address to the trace stack on thread info.*/
static int push_return_trace(unsigned long ret, unsigned long long time,
unsigned long func, int *depth)
{
int index;
if (!current->ret_stack)
return -EBUSY;
/* The return trace stack is full */
if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
atomic_inc(&current->trace_overrun);
return -EBUSY;
}
index = ++current->curr_ret_stack;
barrier();
current->ret_stack[index].ret = ret;
current->ret_stack[index].func = func;
current->ret_stack[index].calltime = time;
*depth = index;
return 0;
}
/* Retrieve a function return address to the trace stack on thread info.*/
static void pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret)
{
int index;
index = current->curr_ret_stack;
if (unlikely(index < 0)) {
ftrace_graph_stop();
WARN_ON(1);
/* Might as well panic, otherwise we have no where to go */
*ret = (unsigned long)
dereference_function_descriptor(&panic);
return;
}
*ret = current->ret_stack[index].ret;
trace->func = current->ret_stack[index].func;
trace->calltime = current->ret_stack[index].calltime;
trace->overrun = atomic_read(&current->trace_overrun);
trace->depth = index;
barrier();
current->curr_ret_stack--;
}
/*
* Send the trace to the ring-buffer.
* @return the original return address.
*/
unsigned long ftrace_return_to_handler(unsigned long retval0,
unsigned long retval1)
{
struct ftrace_graph_ret trace;
unsigned long ret;
pop_return_trace(&trace, &ret);
trace.rettime = local_clock();
ftrace_graph_return(&trace);
if (unlikely(!ret)) {
ftrace_graph_stop();
WARN_ON(1);
/* Might as well panic. What else to do? */
ret = (unsigned long)
dereference_function_descriptor(&panic);
}
/* HACK: we hand over the old functions' return values
in %r23 and %r24. Assembly in entry.S will take care
and move those to their final registers %ret0 and %ret1 */
asm( "copy %0, %%r23 \n\t"
"copy %1, %%r24 \n" : : "r" (retval0), "r" (retval1) );
return ret;
}
/*
* Hook the return address and push it in the stack of return addrs
* in current thread info.
*/
void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr)
static void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr)
{
unsigned long old;
unsigned long long calltime;
struct ftrace_graph_ent trace;
extern int parisc_return_to_handler;
if (unlikely(ftrace_graph_is_dead()))
return;
@ -119,64 +36,47 @@ void prepare_ftrace_return(unsigned long *parent, unsigned long self_addr)
return;
old = *parent;
*parent = (unsigned long)
dereference_function_descriptor(&return_to_handler);
if (unlikely(!__kernel_text_address(old))) {
ftrace_graph_stop();
*parent = old;
WARN_ON(1);
return;
}
calltime = local_clock();
if (push_return_trace(old, calltime,
self_addr, &trace.depth) == -EBUSY) {
*parent = old;
return;
}
trace.func = self_addr;
trace.depth = current->curr_ret_stack + 1;
/* Only trace if the calling function expects to */
if (!ftrace_graph_entry(&trace)) {
current->curr_ret_stack--;
*parent = old;
}
}
if (!ftrace_graph_entry(&trace))
return;
if (ftrace_push_return_trace(old, self_addr, &trace.depth,
0 ) == -EBUSY)
return;
/* activate parisc_return_to_handler() as return point */
*parent = (unsigned long) &parisc_return_to_handler;
}
#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
void ftrace_function_trampoline(unsigned long parent,
void notrace ftrace_function_trampoline(unsigned long parent,
unsigned long self_addr,
unsigned long org_sp_gr3)
{
extern ftrace_func_t ftrace_trace_function;
extern ftrace_func_t ftrace_trace_function; /* depends on CONFIG_DYNAMIC_FTRACE */
extern int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace);
if (ftrace_trace_function != ftrace_stub) {
ftrace_trace_function(parent, self_addr);
/* struct ftrace_ops *op, struct pt_regs *regs); */
ftrace_trace_function(parent, self_addr, NULL, NULL);
return;
}
#ifdef CONFIG_FUNCTION_GRAPH_TRACER
if (ftrace_graph_entry && ftrace_graph_return) {
unsigned long sp;
if (ftrace_graph_return != (trace_func_graph_ret_t) ftrace_stub ||
ftrace_graph_entry != ftrace_graph_entry_stub) {
unsigned long *parent_rp;
asm volatile ("copy %%r30, %0" : "=r"(sp));
/* sanity check: is stack pointer which we got from
assembler function in entry.S in a reasonable
range compared to current stack pointer? */
if ((sp - org_sp_gr3) > 0x400)
return;
/* calculate pointer to %rp in stack */
parent_rp = (unsigned long *) org_sp_gr3 - 0x10;
parent_rp = (unsigned long *) (org_sp_gr3 - RP_OFFSET);
/* sanity check: parent_rp should hold parent */
if (*parent_rp != parent)
return;
prepare_ftrace_return(parent_rp, self_addr);
return;
}

View File

@ -129,6 +129,15 @@ $pgt_fill_loop:
/* And the stack pointer too */
ldo THREAD_SZ_ALGN(%r6),%sp
#if defined(CONFIG_64BIT) && defined(CONFIG_FUNCTION_TRACER)
.import _mcount,data
/* initialize mcount FPTR */
/* Get the global data pointer */
loadgp
load32 PA(_mcount), %r10
std %dp,0x18(%r10)
#endif
#ifdef CONFIG_SMP
/* Set the smp rendezvous address into page zero.
** It would be safer to do this in init_smp_config() but