Blackfin: add support for dynamic ftrace

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

arch/blackfin/Kconfig

@@ -25,6 +25,8 @@ config BLACKFIN
 	def_bool y
 	select HAVE_ARCH_KGDB
 	select HAVE_ARCH_TRACEHOOK
+	select HAVE_DYNAMIC_FTRACE
+	select HAVE_FTRACE_MCOUNT_RECORD
 	select HAVE_FUNCTION_GRAPH_TRACER
 	select HAVE_FUNCTION_TRACER
 	select HAVE_FUNCTION_TRACE_MCOUNT_TEST

arch/blackfin/include/asm/ftrace.h

@@ -12,6 +12,22 @@
 
 #ifndef __ASSEMBLY__
 
+#ifdef CONFIG_DYNAMIC_FTRACE
+
+extern void _mcount(void);
+#define MCOUNT_ADDR ((unsigned long)_mcount)
+
+static inline unsigned long ftrace_call_adjust(unsigned long addr)
+{
+	return addr;
+}
+
+struct dyn_arch_ftrace {
+	/* No extra data needed for Blackfin */
+};
+
+#endif
+
 #ifdef CONFIG_FRAME_POINTER
 #include <linux/mm.h>
 

arch/blackfin/kernel/Makefile

@@ -16,6 +16,7 @@ else
     obj-y += time.o
 endif
 
+obj-$(CONFIG_DYNAMIC_FTRACE)         += ftrace.o
 obj-$(CONFIG_FUNCTION_TRACER)        += ftrace-entry.o
 obj-$(CONFIG_FUNCTION_GRAPH_TRACER)  += ftrace.o
 CFLAGS_REMOVE_ftrace.o = -pg

arch/blackfin/kernel/ftrace-entry.S

@@ -10,6 +10,18 @@
 
 .text
 
+#ifdef CONFIG_DYNAMIC_FTRACE
+
+/* Simple stub so we can boot the kernel until runtime patching has
+ * disabled all calls to this.  Then it'll be unused.
+ */
+ENTRY(__mcount)
+# if ANOMALY_05000371
+	nop; nop; nop; nop;
+# endif
+	rts;
+ENDPROC(__mcount)
+
 /* GCC will have called us before setting up the function prologue, so we
  * can clobber the normal scratch registers, but we need to make sure to
  * save/restore the registers used for argument passing (R0-R2) in case
@@ -20,6 +32,56 @@
  * function.  And since GCC pushed the previous RETS for us, the previous
  * function will be waiting there.  mmmm pie.
  */
+ENTRY(_ftrace_caller)
+# ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
+	/* optional micro optimization: return if stopped */
+	p1.l = _function_trace_stop;
+	p1.h = _function_trace_stop;
+	r3 = [p1];
+	cc = r3 == 0;
+	if ! cc jump _ftrace_stub (bp);
+# endif
+
+	/* save first/second/third function arg and the return register */
+	[--sp] = r2;
+	[--sp] = r0;
+	[--sp] = r1;
+	[--sp] = rets;
+
+	/* function_trace_call(unsigned long ip, unsigned long parent_ip):
+	 *  ip: this point was called by ...
+	 *  parent_ip: ... this function
+	 * the ip itself will need adjusting for the mcount call
+	 */
+	r0 = rets;
+	r1 = [sp + 16];	/* skip the 4 local regs on stack */
+	r0 += -MCOUNT_INSN_SIZE;
+
+.globl _ftrace_call
+_ftrace_call:
+	call _ftrace_stub
+
+# ifdef CONFIG_FUNCTION_GRAPH_TRACER
+.globl _ftrace_graph_call
+_ftrace_graph_call:
+	nop;	/* jump _ftrace_graph_caller; */
+# endif
+
+	/* restore state and get out of dodge */
+.Lfinish_trace:
+	rets = [sp++];
+	r1 = [sp++];
+	r0 = [sp++];
+	r2 = [sp++];
+
+.globl _ftrace_stub
+_ftrace_stub:
+	rts;
+ENDPROC(_ftrace_caller)
+
+#else
+
+/* See documentation for _ftrace_caller */
 ENTRY(__mcount)
 # ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
 	/* optional micro optimization: return if stopped */
@@ -103,6 +165,8 @@ _ftrace_stub:
 	rts;
 ENDPROC(__mcount)
 
+#endif
+
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 /* The prepare_ftrace_return() function is similar to the trace function
  * except it takes a pointer to the location of the frompc.  This is so
@@ -110,6 +174,7 @@ ENDPROC(__mcount)
  * purposes.
  */
 ENTRY(_ftrace_graph_caller)
+# ifndef CONFIG_DYNAMIC_FTRACE
 	/* save first/second function arg and the return register */
 	[--sp] = r0;
 	[--sp] = r1;
@@ -118,6 +183,10 @@ ENTRY(_ftrace_graph_caller)
 	/* prepare_ftrace_return(parent, self_addr, frame_pointer) */
 	r0 = sp;	/* unsigned long *parent */
 	r1 = rets;	/* unsigned long self_addr */
+# else
+	r0 = sp;	/* unsigned long *parent */
+	r1 = [sp];	/* unsigned long self_addr */
+# endif
 # ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
 	r2 = fp;	/* unsigned long frame_pointer */
 # endif

arch/blackfin/kernel/ftrace.c

@@ -1,17 +1,101 @@
 /*
  * ftrace graph code
  *
- * Copyright (C) 2009 Analog Devices Inc.
+ * Copyright (C) 2009-2010 Analog Devices Inc.
  * Licensed under the GPL-2 or later.
  */
 
 #include <linux/ftrace.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
+#include <linux/uaccess.h>
 #include <asm/atomic.h>
+#include <asm/cacheflush.h>
+
+#ifdef CONFIG_DYNAMIC_FTRACE
+
+static const unsigned char mnop[] = {
+	0x03, 0xc0, 0x00, 0x18, /* MNOP; */
+	0x03, 0xc0, 0x00, 0x18, /* MNOP; */
+};
+
+static void bfin_make_pcrel24(unsigned char *insn, unsigned long src,
+                              unsigned long dst)
+{
+	uint32_t pcrel = (dst - src) >> 1;
+	insn[0] = pcrel >> 16;
+	insn[1] = 0xe3;
+	insn[2] = pcrel;
+	insn[3] = pcrel >> 8;
+}
+#define bfin_make_pcrel24(insn, src, dst) bfin_make_pcrel24(insn, src, (unsigned long)(dst))
+
+static int ftrace_modify_code(unsigned long ip, const unsigned char *code,
+                              unsigned long len)
+{
+	int ret = probe_kernel_write((void *)ip, (void *)code, len);
+	flush_icache_range(ip, ip + len);
+	return ret;
+}
+
+int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec,
+                    unsigned long addr)
+{
+	/* Turn the mcount call site into two MNOPs as those are 32bit insns */
+	return ftrace_modify_code(rec->ip, mnop, sizeof(mnop));
+}
+
+int ftrace_make_call(struct dyn_ftrace *rec, unsigned long addr)
+{
+	/* Restore the mcount call site */
+	unsigned char call[8];
+	call[0] = 0x67; /* [--SP] = RETS; */
+	call[1] = 0x01;
+	bfin_make_pcrel24(&call[2], rec->ip + 2, addr);
+	call[6] = 0x27; /* RETS = [SP++]; */
+	call[7] = 0x01;
+	return ftrace_modify_code(rec->ip, call, sizeof(call));
+}
+
+int ftrace_update_ftrace_func(ftrace_func_t func)
+{
+	unsigned char call[4];
+	unsigned long ip = (unsigned long)&ftrace_call;
+	bfin_make_pcrel24(call, ip, func);
+	return ftrace_modify_code(ip, call, sizeof(call));
+}
+
+int __init ftrace_dyn_arch_init(void *data)
+{
+	/* return value is done indirectly via data */
+	*(unsigned long *)data = 0;
+
+	return 0;
+}
+
+#endif
 
 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
 
+# ifdef CONFIG_DYNAMIC_FTRACE
+
+extern void ftrace_graph_call(void);
+
+int ftrace_enable_ftrace_graph_caller(void)
+{
+	unsigned long ip = (unsigned long)&ftrace_graph_call;
+	uint16_t jump_pcrel12 = ((unsigned long)&ftrace_graph_caller - ip) >> 1;
+	jump_pcrel12 |= 0x2000;
+	return ftrace_modify_code(ip, (void *)&jump_pcrel12, sizeof(jump_pcrel12));
+}
+
+int ftrace_disable_ftrace_graph_caller(void)
+{
+	return ftrace_modify_code((unsigned long)&ftrace_graph_call, empty_zero_page, 2);
+}
+
+# endif
+
 /*
  * Hook the return address and push it in the stack of return addrs
  * in current thread info.