sh: add sleazy FPU optimization

sh port of the sLeAZY-fpu feature currently implemented for some architectures
such us i386.

Right now the SH kernel has a 100% lazy fpu behaviour.
This is of course great for applications that have very sporadic or no FPU use.
However for very frequent FPU users...  you take an extra trap every context
switch.
The patch below adds a simple heuristic to this code: after 5 consecutive
context switches of FPU use, the lazy behavior is disabled and the context
gets restored every context switch.
After 256 switches, this is reset and the 100% lazy behavior is returned.

Tests with LMbench showed no regression.
I saw a little improvement due to the prefetching (~2%).

The tests below also show that, with this sLeazy patch, indeed,
the number of FPU exceptions is reduced.
To test this. I hacked the lat_ctx LMBench to use the FPU a little more.

   sLeasy implementation
   ===========================================
   switch_to calls            |  79326
   sleasy   calls             |  42577
   do_fpu_state_restore  calls|  59232
   restore_fpu   calls        |  59032

   Exceptions:  0x800 (FPU disabled  ): 16604

   100% Leazy (default implementation)
   ===========================================
   switch_to  calls            |  79690
   do_fpu_state_restore calls  |  53299
   restore_fpu  calls          |   53101

   Exceptions: 0x800 (FPU disabled  ):  53273
Signed-off-by: Giuseppe Cavallaro <peppe.cavallaro@st.com>
Signed-off-by: Stuart Menefy <stuart.menefy@st.com>
Signed-off-by: Paul Mundt <lethal@linux-sh.org>

arch/sh/include/asm/fpu.h

@@ -19,6 +19,7 @@ static inline void grab_fpu(struct pt_regs *regs)
 struct task_struct;
 
 extern void save_fpu(struct task_struct *__tsk, struct pt_regs *regs);
+void fpu_state_restore(struct pt_regs *regs);
 #else
 
 #define release_fpu(regs)	do { } while (0)
@@ -44,6 +45,8 @@ static inline void unlazy_fpu(struct task_struct *tsk, struct pt_regs *regs)
 	preempt_disable();
 	if (test_tsk_thread_flag(tsk, TIF_USEDFPU))
 		save_fpu(tsk, regs);
+	else
+		tsk->fpu_counter = 0;
 	preempt_enable();
 }
 

arch/sh/kernel/cpu/sh4/fpu.c

@@ -483,18 +483,18 @@ BUILD_TRAP_HANDLER(fpu_error)
 	force_sig(SIGFPE, tsk);
 }
 
-BUILD_TRAP_HANDLER(fpu_state_restore)
+void fpu_state_restore(struct pt_regs *regs)
 {
 	struct task_struct *tsk = current;
-	TRAP_HANDLER_DECL;
 
 	grab_fpu(regs);
-	if (!user_mode(regs)) {
+	if (unlikely(!user_mode(regs))) {
 		printk(KERN_ERR "BUG: FPU is used in kernel mode.\n");
+		BUG();
 		return;
 	}
 
-	if (used_math()) {
+	if (likely(used_math())) {
 		/* Using the FPU again.  */
 		restore_fpu(tsk);
 	} else {
@@ -503,4 +503,12 @@ BUILD_TRAP_HANDLER(fpu_state_restore)
 		set_used_math();
 	}
 	set_tsk_thread_flag(tsk, TIF_USEDFPU);
+	tsk->fpu_counter++;
+}
+
+BUILD_TRAP_HANDLER(fpu_state_restore)
+{
+	TRAP_HANDLER_DECL;
+
+	fpu_state_restore(regs);
 }

arch/sh/kernel/process_32.c

@@ -288,8 +288,14 @@ static void ubc_set_tracing(int asid, unsigned long pc)
 __notrace_funcgraph struct task_struct *
 __switch_to(struct task_struct *prev, struct task_struct *next)
 {
+	struct thread_struct *next_t = &next->thread;
+
 #if defined(CONFIG_SH_FPU)
 	unlazy_fpu(prev, task_pt_regs(prev));
+
+	/* we're going to use this soon, after a few expensive things */
+	if (next->fpu_counter > 5)
+		prefetch(&next_t->fpu.hard);
 #endif
 
 #ifdef CONFIG_MMU
@@ -321,6 +327,16 @@ __switch_to(struct task_struct *prev, struct task_struct *next)
 #endif
 	}
 
+#if defined(CONFIG_SH_FPU)
+	/* If the task has used fpu the last 5 timeslices, just do a full
+	 * restore of the math state immediately to avoid the trap; the
+	 * chances of needing FPU soon are obviously high now
+	 */
+	if (next->fpu_counter > 5) {
+		fpu_state_restore(task_pt_regs(next));
+	}
+#endif
+
 	return prev;
 }