SMP FPU management: similar to what sparc does, no lazy fpu context

switching in SMP mode, use PF_USEDFPU to determine whether a program
used the fpu in the last time quantum and so needs the fpu context
to be saved during context switch.

7 files changed, 61 insertions, 15 deletions

diff --git a/arch/mips64/kernel/process.c b/arch/mips64/kernel/process.c
index d6d8b7435..22ce0ed6d 100644
--- a/arch/mips64/kernel/process.c
+++ b/arch/mips64/kernel/process.c
@@ -7,6 +7,7 @@
  * Copyright (C) 1994 - 1999 by Ralf Baechle and others.
  * Copyright (C) 1999 Silicon Graphics, Inc.
  */
+#include <linux/config.h>
 #include <linux/errno.h>
 #include <linux/sched.h>
 #include <linux/kernel.h>
@@ -52,20 +53,20 @@ asmlinkage void ret_from_fork(void);
 void exit_thread(void)
 {
 	/* Forget lazy fpu state */
-	if (last_task_used_math == current) {
+	if (IS_FPU_OWNER()) {
 		set_cp0_status(ST0_CU1, ST0_CU1);
 		__asm__ __volatile__("cfc1\t$0,$31");
-		last_task_used_math = NULL;
+		CLEAR_FPU_OWNER();
 	}
 }
 
 void flush_thread(void)
 {
 	/* Forget lazy fpu state */
-	if (last_task_used_math == current) {
+	if (IS_FPU_OWNER()) {
 		set_cp0_status(ST0_CU1, ST0_CU1);
 		__asm__ __volatile__("cfc1\t$0,$31");
-		last_task_used_math = NULL;
+		CLEAR_FPU_OWNER();
 	}
 }
 
@@ -77,7 +78,7 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 
 	childksp = (unsigned long)p + KERNEL_STACK_SIZE - 32;
 
-	if (last_task_used_math == current) {
+	if (IS_FPU_OWNER()) {
 		save_fp(p);
 	}
 	/* set up new TSS. */
diff --git a/arch/mips64/kernel/ptrace.c b/arch/mips64/kernel/ptrace.c
index 0301a66a3..b79a61f0e 100644
--- a/arch/mips64/kernel/ptrace.c
+++ b/arch/mips64/kernel/ptrace.c
@@ -13,6 +13,7 @@
  * At this time Linux/MIPS64 only supports syscall tracing, even for 32-bit
  * binaries.
  */
+#include <linux/config.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
@@ -36,7 +37,6 @@ asmlinkage int sys32_ptrace(int request, int pid, int addr, int data)
 	struct task_struct *child;
 	unsigned long flags;
 	int ret;
-	extern void save_fp(void*);
 
 	lock_kernel();
 #if 0
@@ -132,12 +132,14 @@ asmlinkage int sys32_ptrace(int request, int pid, int addr, int data)
 			break;
 		case FPR_BASE ... FPR_BASE + 31:
 			if (child->used_math) {
+#ifndef CONFIG_SMP
 				if (last_task_used_math == child) {
 					set_cp0_status(ST0_CU1, ST0_CU1);
 					save_fp(child);
 					set_cp0_status(ST0_CU1, 0);
 					last_task_used_math = NULL;
 				}
+#endif
 				tmp = child->thread.fpu.hard.fp_regs[addr - 32];
 			} else {
 				tmp = -EIO;
@@ -199,6 +201,7 @@ asmlinkage int sys32_ptrace(int request, int pid, int addr, int data)
 		case FPR_BASE ... FPR_BASE + 31: {
 			unsigned long *fregs;
 			if (child->used_math) {
+#ifndef CONFIG_SMP
 				if (last_task_used_math == child) {
 					set_cp0_status(ST0_CU1, ST0_CU1);
 					save_fp(child);
@@ -206,6 +209,7 @@ asmlinkage int sys32_ptrace(int request, int pid, int addr, int data)
 					last_task_used_math = NULL;
 					regs->cp0_status &= ~ST0_CU1;
 				}
+#endif
 			} else {
 				/* FP not yet used  */
 				memset(&child->thread.fpu.hard, ~0,
@@ -293,7 +297,6 @@ asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
 	struct task_struct *child;
 	unsigned long flags;
 	int ret;
-	extern void save_fp(void*);
 
 	lock_kernel();
 #if 0
@@ -389,12 +392,14 @@ asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
 			break;
 		case FPR_BASE ... FPR_BASE + 31:
 			if (child->used_math) {
+#ifndef CONFIG_SMP
 				if (last_task_used_math == child) {
 					set_cp0_status(ST0_CU1, ST0_CU1);
 					save_fp(child);
 					set_cp0_status(ST0_CU1, 0);
 					last_task_used_math = NULL;
 				}
+#endif
 				tmp = child->thread.fpu.hard.fp_regs[addr - 32];
 			} else {
 				tmp = -EIO;
@@ -456,6 +461,7 @@ asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
 		case FPR_BASE ... FPR_BASE + 31: {
 			unsigned long *fregs;
 			if (child->used_math) {
+#ifndef CONFIG_SMP
 				if (last_task_used_math == child) {
 					set_cp0_status(ST0_CU1, ST0_CU1);
 					save_fp(child);
@@ -463,6 +469,7 @@ asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
 					last_task_used_math = NULL;
 					regs->cp0_status &= ~ST0_CU1;
 				}
+#endif
 			} else {
 				/* FP not yet used  */
 				memset(&child->thread.fpu.hard, ~0,
diff --git a/arch/mips64/kernel/signal.c b/arch/mips64/kernel/signal.c
index 40a727831..7f644a437 100644
--- a/arch/mips64/kernel/signal.c
+++ b/arch/mips64/kernel/signal.c
@@ -26,6 +26,7 @@
 #include <asm/stackframe.h>
 #include <asm/uaccess.h>
 #include <asm/ucontext.h>
+#include <asm/system.h>
 
 #define DEBUG_SIG 0
 
@@ -209,8 +210,8 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext *sc)
 
 	err |= __get_user(owned_fp, &sc->sc_ownedfp);
 	if (owned_fp) {
-		if (current == last_task_used_math) {
-			last_task_used_math = 0;
+		if (IS_FPU_OWNER()) {
+			CLEAR_FPU_OWNER();
 			regs->cp0_status &= ~ST0_CU1;
 		}
 		current->used_math = 1;
@@ -338,9 +339,9 @@ setup_sigcontext(struct pt_regs *regs, struct sigcontext *sc)
 	err |= __put_user(regs->cp0_badvaddr, &sc->sc_badvaddr);
 
 	if (current->used_math) {	/* fp is active.  */
-		if (current == last_task_used_math) {
+		if (IS_FPU_OWNER()) {
 			lazy_fpu_switch(current, 0);
-			last_task_used_math = NULL;
+			CLEAR_FPU_OWNER();
 			regs->cp0_status &= ~ST0_CU1;
 		}
 		err |= __put_user(1, &sc->sc_ownedfp);
diff --git a/arch/mips64/kernel/signal32.c b/arch/mips64/kernel/signal32.c
index 4addc707c..e33a69d5e 100644
--- a/arch/mips64/kernel/signal32.c
+++ b/arch/mips64/kernel/signal32.c
@@ -8,6 +8,7 @@
  * Copyright (C) 1994 - 1999  Ralf Baechle
  * Copyright (C) 1999 Silicon Graphics, Inc.
  */
+#include <linux/config.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
 #include <linux/smp.h>
@@ -25,6 +26,7 @@
 #include <asm/stackframe.h>
 #include <asm/uaccess.h>
 #include <asm/ucontext.h>
+#include <asm/system.h>
 
 #define DEBUG_SIG 0
 
@@ -239,8 +241,8 @@ restore_sigcontext(struct pt_regs *regs, struct sigcontext *sc)
 
 	err |= __get_user(owned_fp, &sc->sc_ownedfp);
 	if (owned_fp) {
-		if (current == last_task_used_math) {
-			last_task_used_math = 0;
+		if (IS_FPU_OWNER()) {
+			CLEAR_FPU_OWNER();
 			regs->cp0_status &= ~ST0_CU1;
 		}
 		current->used_math = 1;
@@ -373,9 +375,9 @@ setup_sigcontext(struct pt_regs *regs, struct sigcontext *sc)
 	err |= __put_user(regs->cp0_badvaddr, &sc->sc_badvaddr);
 
 	if (current->used_math) {	/* fp is active.  */
-		if (current == last_task_used_math) {
+		if (IS_FPU_OWNER()) {
 			lazy_fpu_switch(current, 0);
-			last_task_used_math = NULL;
+			CLEAR_FPU_OWNER();
 			regs->cp0_status &= ~ST0_CU1;
 		}
 		err |= __put_user(1, &sc->sc_ownedfp);
diff --git a/arch/mips64/kernel/traps.c b/arch/mips64/kernel/traps.c
index 9f97fb0d8..fa58017cf 100644
--- a/arch/mips64/kernel/traps.c
+++ b/arch/mips64/kernel/traps.c
@@ -346,6 +346,7 @@ void do_cpu(struct pt_regs *regs)
 		goto bad_cid;
 
 	regs->cp0_status |= ST0_CU1;
+#ifndef CONFIG_SMP
 	if (last_task_used_math == current)
 		return;
 
@@ -357,6 +358,15 @@ void do_cpu(struct pt_regs *regs)
 		current->used_math = 1;
 	}
 	last_task_used_math = current;
+#else
+	if (current->used_math) {
+		lazy_fpu_switch(0, current);
+	} else {
+		init_fpu();
+		current->used_math = 1;
+	}
+	current->flags |= PF_USEDFPU;
+#endif
 	return;
 
 bad_cid:
diff --git a/include/asm-mips64/processor.h b/include/asm-mips64/processor.h
index 0532dbf0c..1932672e7 100644
--- a/include/asm-mips64/processor.h
+++ b/include/asm-mips64/processor.h
@@ -87,6 +87,14 @@ extern int EISA_bus;
 /* Lazy FPU handling on uni-processor */
 extern struct task_struct *last_task_used_math;
 
+#ifndef CONFIG_SMP
+#define IS_FPU_OWNER()		(last_task_used_math == current)
+#define CLEAR_FPU_OWNER()	last_task_used_math = NULL;
+#else
+#define IS_FPU_OWNER()		(current->flags & PF_USEDFPU)
+#define CLEAR_FPU_OWNER()	current->flags &= ~PF_USEDFPU;
+#endif
+
 /*
  * User space process size: 1TB. This is hardcoded into a few places,
  * so don't change it unless you know what you are doing.  TASK_SIZE
diff --git a/include/asm-mips64/system.h b/include/asm-mips64/system.h
index 7f518a8b8..3592b3821 100644
--- a/include/asm-mips64/system.h
+++ b/include/asm-mips64/system.h
@@ -169,8 +169,25 @@ extern asmlinkage void *resume(void *last, void *next);
 #endif /* !defined (_LANGUAGE_ASSEMBLY) */
 
 #define prepare_to_switch()	do { } while(0)
+
+extern asmlinkage void lazy_fpu_switch(void *, void *);
+extern asmlinkage void init_fpu(void);
+extern asmlinkage void save_fp(void *);
+
+#ifdef CONFIG_SMP
+#define SWITCH_DO_LAZY_FPU \
+	if (prev->flags & PF_USEDFPU) { \
+		lazy_fpu_switch(prev, 0); \
+		set_cp0_status(ST0_CU1, ~ST0_CU1); \
+		prev->flags &= ~PF_USEDFPU; \
+	}
+#else /* CONFIG_SMP */
+#define SWITCH_DO_LAZY_FPU	do { } while(0)
+#endif /* CONFIG_SMP */
+
 #define switch_to(prev,next,last) \
 do { \
+	SWITCH_DO_LAZY_FPU; \
 	(last) = resume(prev, next); \
 } while(0)