o Merge with Linux 2.1.91.

 o First round of bugfixes for the SC/MC CPUs.
 o FPU context switch fixes.
 o Lazy context switches.
 o Faster syscalls.
 o Removed dead code.
 o Shitloads of other things I forgot ...

1 files changed, 129 insertions, 30 deletions

diff --git a/arch/mips/kernel/r4k_switch.S b/arch/mips/kernel/r4k_switch.S
index cb7f9891f..765de0d85 100644
--- a/arch/mips/kernel/r4k_switch.S
+++ b/arch/mips/kernel/r4k_switch.S
@@ -5,6 +5,8 @@
  *
  * Multi-cpu abstraction and macros for easier reading:
  * Copyright (C) 1996 David S. Miller (dm@engr.sgi.com)
+ *
+ * $Id: r4k_switch.S,v 1.3 1998/04/04 13:59:38 ralf Exp $
  */
 #include <asm/asm.h>
 #include <asm/bootinfo.h>
@@ -13,6 +15,7 @@
 #include <asm/fpregdef.h>
 #include <asm/mipsconfig.h>
 #include <asm/mipsregs.h>
+#include <asm/offset.h>
 #include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/processor.h>
@@ -25,43 +28,139 @@
 	.set	mips3
 	.align	5
 	LEAF(r4xx0_resume)
-	mfc0	t1, CP0_STATUS
-	sw	t1, THREAD_STATUS($28)
-	ori	t2, t1, 0x1f
-	xori	t2, t2, 0x1e
-	mtc0	t2, CP0_STATUS
+	mfc0	t1, CP0_STATUS			# fp exception boundary
+	sll	t0, t1, 2
+	bgez	t0, 1f
+	 nop
+	cfc1	zero, fcr31
+1:	sw	t1, THREAD_STATUS($28)
 	CPU_SAVE_NONSCRATCH($28)
-	sll	t2, t1, 2		# Save floating point state
-	bgez	t2, 2f
-	 sw	ra, THREAD_REG31($28)
-	sll	t2, t1, 5
-	bgez	t2, 1f
-	 swc1	$f0, (THREAD_FPU + 0x00)($28)
-	FPU_SAVE_16ODD($28)
-1:
-	FPU_SAVE_16EVEN($28, t1)		# clobbers t1
-2:
+	sw	ra, THREAD_REG31($28)
+
+	/*
+	 * The order of restoring the registers takes care of the race
+	 * updating $28, $29 and kernelsp without disabling ints.
+	 */
 	move	$28, a0
+	CPU_RESTORE_NONSCRATCH($28)
+	addiu	t0, $28, KERNEL_STACK_SIZE-32
+	sw	t0, kernelsp
 	lw	a3, TASK_MM($28)
 	lw	a2, THREAD_STATUS($28)
 	lw	a3, MM_CONTEXT(a3)
-	ori	t1, a2, 1		# restore fpu, pipeline magic
+	mtc0	a2, CP0_STATUS
 	andi	a3, a3, 0xff
-	xori	t1, t1, 1
-	mtc0	a3, CP0_ENTRYHI
-	mtc0	t1, CP0_STATUS
-	sll	t0, a2, 2
-	bgez	t0, 2f
-	 sll	t0, a2, 5
+	jr	ra
+	 mtc0	a3, CP0_ENTRYHI
+	END(r4xx0_resume)
+
+/*
+ * Do lazy fpu context switch.  Saves FPU context to the process in a0
+ * and loads the new context of the current process.
+ */
+
+#define ST_OFF (KERNEL_STACK_SIZE - 32 - PT_SIZE + PT_STATUS)
+
+LEAF(r4xx0_lazy_fpu_switch)
+	mfc0	t0, CP0_STATUS			# enable cp1
+	li	t3, 0x20000000
+	or	t0, t3
+	mtc0	t0, CP0_STATUS
+
+	beqz	a0, 2f				# Save floating point state
+	 nor	t3, zero, t3
+	lw	t1, ST_OFF(a0)			# last thread looses fpu
+	and	t1, t3
+	sw	t1, ST_OFF(a0)
+	sll	t2, t1, 5
+	bgez	t2, 1f
+	 sdc1	$f0, (THREAD_FPU + 0x00)(a0)
+	FPU_SAVE_16ODD(a0)
+1:
+	FPU_SAVE_16EVEN(a0, t1)			# clobbers t1
+2:
+
+	sll	t0, t0, 5			# load new fp state
 	bgez	t0, 1f
-	 lwc1	$f0, (THREAD_FPU + 0x00)($28)
+	 ldc1	$f0, (THREAD_FPU + 0x00)($28)
 	FPU_RESTORE_16ODD($28)
 1:
-	FPU_RESTORE_16EVEN($28, t0)	# clobbers t0
-2:
-	CPU_RESTORE_NONSCRATCH($28)
-	lw	t0, THREAD_KSP($28)
-	sw	t0, kernelsp
+	.set	reorder
+	FPU_RESTORE_16EVEN($28, t0)		# clobbers t0
 	jr	ra
-	 mtc0	a2, CP0_STATUS
-	END(r4xx0_resume)
+	END(r4xx0_lazy_fpu_switch)
+
+/*
+ * Save a thread's fp context.
+ */
+	.set	noreorder
+LEAF(r4xx0_save_fp)
+	mfc0	t0, CP0_STATUS
+	sll	t1, t0, 5
+	bgez	t1, 1f				# 16 register mode?
+	 nop
+	FPU_SAVE_16ODD(a0)
+1:
+	FPU_SAVE_16EVEN(a0, t1)			# clobbers t1
+	jr	ra
+	 sdc1	$f0, (THREAD_FPU + 0x00)(a0)
+	END(r4xx0_save_fp)
+
+/*
+ * Load the FPU with signalling NANS.  This bit pattern we're using has
+ * the property that no matter wether considered as single or as double
+ * precission represents signaling NANS.
+ *
+ * We initialize fcr31 to rounding to nearest, no exceptions.
+ */
+
+#define FPU_DEFAULT  0x00000600
+
+LEAF(r4xx0_init_fpu)
+	mfc0	t0, CP0_STATUS
+	li	t1, 0x20000000
+	or	t0, t1
+	mtc0	t0, CP0_STATUS
+	sll	t0, t0, 5
+
+	li	t1, FPU_DEFAULT
+	ctc1	t1, fcr31
+
+	bgez	t0, 1f			# 16 / 32 register mode?
+	 li	t0, -1
+
+	dmtc1	t0, $f1
+	dmtc1	t0, $f3
+	dmtc1	t0, $f5
+	dmtc1	t0, $f7
+	dmtc1	t0, $f9
+	dmtc1	t0, $f11
+	dmtc1	t0, $f13
+	dmtc1	t0, $f15
+	dmtc1	t0, $f17
+	dmtc1	t0, $f19
+	dmtc1	t0, $f21
+	dmtc1	t0, $f23
+	dmtc1	t0, $f25
+	dmtc1	t0, $f27
+	dmtc1	t0, $f29
+	dmtc1	t0, $f31
+
+1:	dmtc1	t0, $f0
+	dmtc1	t0, $f2
+	dmtc1	t0, $f4
+	dmtc1	t0, $f6
+	dmtc1	t0, $f8
+	dmtc1	t0, $f10
+	dmtc1	t0, $f12
+	dmtc1	t0, $f14
+	dmtc1	t0, $f16
+	dmtc1	t0, $f18
+	dmtc1	t0, $f20
+	dmtc1	t0, $f22
+	dmtc1	t0, $f24
+	dmtc1	t0, $f26
+	dmtc1	t0, $f28
+	jr	ra
+	 dmtc1	t0, $f30
+	END(r4xx0_init_fpu)