[SPARC64]: Refine register window trap handling.

When saving and restoing trap state, do the window spill/fill
handling inline so that we never trap deeper than 2 trap levels.
This is important for chips like Niagara.

The window fixup code is massively simplified, and many more
improvements are now possible.

Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 233 insertions, 1 deletions

diff --git a/include/asm-sparc64/ttable.h b/include/asm-sparc64/ttable.h
index f557db4faf84..f912f52c0c7f 100644
--- a/include/asm-sparc64/ttable.h
+++ b/include/asm-sparc64/ttable.h
@@ -93,7 +93,7 @@
 	
 #define SYSCALL_TRAP(routine, systbl)			\
 	sethi	%hi(109f), %g7;				\
-	ba,pt	%xcc, scetrap;				\
+	ba,pt	%xcc, etrap;				\
 109:	 or	%g7, %lo(109b), %g7;			\
 	sethi	%hi(systbl), %l7;			\
 	ba,pt	%xcc, routine;				\
@@ -219,6 +219,31 @@
 	saved; retry; nop; nop; nop; nop; nop; nop;	\
 	nop; nop; nop; nop; nop; nop; nop; nop;
 
+#define SPILL_0_NORMAL_ETRAP				\
+etrap_kernel_spill:					\
+	stx	%l0, [%sp + STACK_BIAS + 0x00];		\
+	stx	%l1, [%sp + STACK_BIAS + 0x08];		\
+	stx	%l2, [%sp + STACK_BIAS + 0x10];		\
+	stx	%l3, [%sp + STACK_BIAS + 0x18];		\
+	stx	%l4, [%sp + STACK_BIAS + 0x20];		\
+	stx	%l5, [%sp + STACK_BIAS + 0x28];		\
+	stx	%l6, [%sp + STACK_BIAS + 0x30];		\
+	stx	%l7, [%sp + STACK_BIAS + 0x38];		\
+	stx	%i0, [%sp + STACK_BIAS + 0x40];		\
+	stx	%i1, [%sp + STACK_BIAS + 0x48];		\
+	stx	%i2, [%sp + STACK_BIAS + 0x50];		\
+	stx	%i3, [%sp + STACK_BIAS + 0x58];		\
+	stx	%i4, [%sp + STACK_BIAS + 0x60];		\
+	stx	%i5, [%sp + STACK_BIAS + 0x68];		\
+	stx	%i6, [%sp + STACK_BIAS + 0x70];		\
+	stx	%i7, [%sp + STACK_BIAS + 0x78];		\
+	saved;						\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	wrpr	%g1, %cwp;				\
+	nop; nop; nop; nop; nop; nop; nop; nop;		\
+	nop; nop; nop; nop;
+
 /* Normal 64bit spill */
 #define SPILL_1_GENERIC(ASI)				\
 	add	%sp, STACK_BIAS + 0x00, %g1;		\
@@ -252,6 +277,67 @@
 	b,a,pt	%xcc, spill_fixup_mna;			\
 	b,a,pt	%xcc, spill_fixup;
 
+#define SPILL_1_GENERIC_ETRAP				\
+etrap_user_spill_64bit:					\
+	stxa	%l0, [%sp + STACK_BIAS + 0x00] %asi;	\
+	stxa	%l1, [%sp + STACK_BIAS + 0x08] %asi;	\
+	stxa	%l2, [%sp + STACK_BIAS + 0x10] %asi;	\
+	stxa	%l3, [%sp + STACK_BIAS + 0x18] %asi;	\
+	stxa	%l4, [%sp + STACK_BIAS + 0x20] %asi;	\
+	stxa	%l5, [%sp + STACK_BIAS + 0x28] %asi;	\
+	stxa	%l6, [%sp + STACK_BIAS + 0x30] %asi;	\
+	stxa	%l7, [%sp + STACK_BIAS + 0x38] %asi;	\
+	stxa	%i0, [%sp + STACK_BIAS + 0x40] %asi;	\
+	stxa	%i1, [%sp + STACK_BIAS + 0x48] %asi;	\
+	stxa	%i2, [%sp + STACK_BIAS + 0x50] %asi;	\
+	stxa	%i3, [%sp + STACK_BIAS + 0x58] %asi;	\
+	stxa	%i4, [%sp + STACK_BIAS + 0x60] %asi;	\
+	stxa	%i5, [%sp + STACK_BIAS + 0x68] %asi;	\
+	stxa	%i6, [%sp + STACK_BIAS + 0x70] %asi;	\
+	stxa	%i7, [%sp + STACK_BIAS + 0x78] %asi;	\
+	saved;						\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop;				\
+	ba,a,pt	%xcc, etrap_spill_fixup_64bit;		\
+	ba,a,pt	%xcc, etrap_spill_fixup_64bit;		\
+	ba,a,pt	%xcc, etrap_spill_fixup_64bit;
+
+#define SPILL_1_GENERIC_ETRAP_FIXUP			\
+etrap_spill_fixup_64bit:				\
+	ldub	[%g6 + TI_WSAVED], %g1;			\
+	sll	%g1, 3, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stx	%sp, [%g3 + TI_RWIN_SPTRS];		\
+	sll	%g1, 7, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stx	%l0, [%g3 + TI_REG_WINDOW + 0x00];	\
+	stx	%l1, [%g3 + TI_REG_WINDOW + 0x08];	\
+	stx	%l2, [%g3 + TI_REG_WINDOW + 0x10];	\
+	stx	%l3, [%g3 + TI_REG_WINDOW + 0x18];	\
+	stx	%l4, [%g3 + TI_REG_WINDOW + 0x20];	\
+	stx	%l5, [%g3 + TI_REG_WINDOW + 0x28];	\
+	stx	%l6, [%g3 + TI_REG_WINDOW + 0x30];	\
+	stx	%l7, [%g3 + TI_REG_WINDOW + 0x38];	\
+	stx	%i0, [%g3 + TI_REG_WINDOW + 0x40];	\
+	stx	%i1, [%g3 + TI_REG_WINDOW + 0x48];	\
+	stx	%i2, [%g3 + TI_REG_WINDOW + 0x50];	\
+	stx	%i3, [%g3 + TI_REG_WINDOW + 0x58];	\
+	stx	%i4, [%g3 + TI_REG_WINDOW + 0x60];	\
+	stx	%i5, [%g3 + TI_REG_WINDOW + 0x68];	\
+	stx	%i6, [%g3 + TI_REG_WINDOW + 0x70];	\
+	stx	%i7, [%g3 + TI_REG_WINDOW + 0x78];	\
+	add	%g1, 1, %g1;				\
+	stb	%g1, [%g6 + TI_WSAVED];			\
+	saved;						\
+	rdpr	%cwp, %g1;				\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop
+
 /* Normal 32bit spill */
 #define SPILL_2_GENERIC(ASI)				\
 	srl	%sp, 0, %sp;				\
@@ -285,6 +371,68 @@
 	b,a,pt	%xcc, spill_fixup_mna;			\
 	b,a,pt	%xcc, spill_fixup;
 
+#define SPILL_2_GENERIC_ETRAP		\
+etrap_user_spill_32bit:			\
+	srl	%sp, 0, %sp;		\
+	stwa	%l0, [%sp + 0x00] %asi;	\
+	stwa	%l1, [%sp + 0x04] %asi;	\
+	stwa	%l2, [%sp + 0x08] %asi;	\
+	stwa	%l3, [%sp + 0x0c] %asi;	\
+	stwa	%l4, [%sp + 0x10] %asi;	\
+	stwa	%l5, [%sp + 0x14] %asi;	\
+	stwa	%l6, [%sp + 0x18] %asi;	\
+	stwa	%l7, [%sp + 0x1c] %asi;	\
+	stwa	%i0, [%sp + 0x20] %asi;	\
+	stwa	%i1, [%sp + 0x24] %asi;	\
+	stwa	%i2, [%sp + 0x28] %asi;	\
+	stwa	%i3, [%sp + 0x2c] %asi;	\
+	stwa	%i4, [%sp + 0x30] %asi;	\
+	stwa	%i5, [%sp + 0x34] %asi;	\
+	stwa	%i6, [%sp + 0x38] %asi;	\
+	stwa	%i7, [%sp + 0x3c] %asi;	\
+	saved;				\
+	sub	%g1, 2, %g1;		\
+	ba,pt	%xcc, etrap_save;	\
+	 wrpr	%g1, %cwp;		\
+	nop; nop; nop; nop;		\
+	nop; nop; nop; nop;		\
+	ba,a,pt	%xcc, etrap_spill_fixup_32bit; \
+	ba,a,pt	%xcc, etrap_spill_fixup_32bit; \
+	ba,a,pt	%xcc, etrap_spill_fixup_32bit;
+
+#define SPILL_2_GENERIC_ETRAP_FIXUP			\
+etrap_spill_fixup_32bit:				\
+	ldub	[%g6 + TI_WSAVED], %g1;			\
+	sll	%g1, 3, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stx	%sp, [%g3 + TI_RWIN_SPTRS];		\
+	sll	%g1, 7, %g3;				\
+	add	%g6, %g3, %g3;				\
+	stw	%l0, [%g3 + TI_REG_WINDOW + 0x00];	\
+	stw	%l1, [%g3 + TI_REG_WINDOW + 0x04];	\
+	stw	%l2, [%g3 + TI_REG_WINDOW + 0x08];	\
+	stw	%l3, [%g3 + TI_REG_WINDOW + 0x0c];	\
+	stw	%l4, [%g3 + TI_REG_WINDOW + 0x10];	\
+	stw	%l5, [%g3 + TI_REG_WINDOW + 0x14];	\
+	stw	%l6, [%g3 + TI_REG_WINDOW + 0x18];	\
+	stw	%l7, [%g3 + TI_REG_WINDOW + 0x1c];	\
+	stw	%i0, [%g3 + TI_REG_WINDOW + 0x20];	\
+	stw	%i1, [%g3 + TI_REG_WINDOW + 0x24];	\
+	stw	%i2, [%g3 + TI_REG_WINDOW + 0x28];	\
+	stw	%i3, [%g3 + TI_REG_WINDOW + 0x2c];	\
+	stw	%i4, [%g3 + TI_REG_WINDOW + 0x30];	\
+	stw	%i5, [%g3 + TI_REG_WINDOW + 0x34];	\
+	stw	%i6, [%g3 + TI_REG_WINDOW + 0x38];	\
+	stw	%i7, [%g3 + TI_REG_WINDOW + 0x3c];	\
+	add	%g1, 1, %g1;				\
+	stb	%g1, [%g6 + TI_WSAVED];			\
+	saved;						\
+	rdpr	%cwp, %g1;				\
+	sub	%g1, 2, %g1;				\
+	ba,pt	%xcc, etrap_save;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop
+
 #define SPILL_1_NORMAL SPILL_1_GENERIC(ASI_AIUP)
 #define SPILL_2_NORMAL SPILL_2_GENERIC(ASI_AIUP)
 #define SPILL_3_NORMAL SPILL_0_NORMAL
@@ -323,6 +471,35 @@
 	restored; retry; nop; nop; nop; nop; nop; nop;	\
 	nop; nop; nop; nop; nop; nop; nop; nop;
 
+#define FILL_0_NORMAL_RTRAP				\
+kern_rtt_fill:						\
+	rdpr	%cwp, %g1;				\
+	sub	%g1, 1, %g1;				\
+	wrpr	%g1, %cwp;				\
+	ldx	[%sp + STACK_BIAS + 0x00], %l0;		\
+	ldx	[%sp + STACK_BIAS + 0x08], %l1;		\
+	ldx	[%sp + STACK_BIAS + 0x10], %l2;		\
+	ldx	[%sp + STACK_BIAS + 0x18], %l3;		\
+	ldx	[%sp + STACK_BIAS + 0x20], %l4;		\
+	ldx	[%sp + STACK_BIAS + 0x28], %l5;		\
+	ldx	[%sp + STACK_BIAS + 0x30], %l6;		\
+	ldx	[%sp + STACK_BIAS + 0x38], %l7;		\
+	ldx	[%sp + STACK_BIAS + 0x40], %i0;		\
+	ldx	[%sp + STACK_BIAS + 0x48], %i1;		\
+	ldx	[%sp + STACK_BIAS + 0x50], %i2;		\
+	ldx	[%sp + STACK_BIAS + 0x58], %i3;		\
+	ldx	[%sp + STACK_BIAS + 0x60], %i4;		\
+	ldx	[%sp + STACK_BIAS + 0x68], %i5;		\
+	ldx	[%sp + STACK_BIAS + 0x70], %i6;		\
+	ldx	[%sp + STACK_BIAS + 0x78], %i7;		\
+	restored;					\
+	add	%g1, 1, %g1;				\
+	ba,pt	%xcc, kern_rtt_restore;			\
+	 wrpr	%g1, %cwp;				\
+	nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop;
+
+
 /* Normal 64bit fill */
 #define FILL_1_GENERIC(ASI)				\
 	add	%sp, STACK_BIAS + 0x00, %g1;		\
@@ -354,6 +531,33 @@
 	b,a,pt	%xcc, fill_fixup_mna;			\
 	b,a,pt	%xcc, fill_fixup;
 
+#define FILL_1_GENERIC_RTRAP				\
+user_rtt_fill_64bit:					\
+	ldxa	[%sp + STACK_BIAS + 0x00] %asi, %l0;	\
+	ldxa	[%sp + STACK_BIAS + 0x08] %asi, %l1;	\
+	ldxa	[%sp + STACK_BIAS + 0x10] %asi, %l2;	\
+	ldxa	[%sp + STACK_BIAS + 0x18] %asi, %l3;	\
+	ldxa	[%sp + STACK_BIAS + 0x20] %asi, %l4;	\
+	ldxa	[%sp + STACK_BIAS + 0x28] %asi, %l5;	\
+	ldxa	[%sp + STACK_BIAS + 0x30] %asi, %l6;	\
+	ldxa	[%sp + STACK_BIAS + 0x38] %asi, %l7;	\
+	ldxa	[%sp + STACK_BIAS + 0x40] %asi, %i0;	\
+	ldxa	[%sp + STACK_BIAS + 0x48] %asi, %i1;	\
+	ldxa	[%sp + STACK_BIAS + 0x50] %asi, %i2;	\
+	ldxa	[%sp + STACK_BIAS + 0x58] %asi, %i3;	\
+	ldxa	[%sp + STACK_BIAS + 0x60] %asi, %i4;	\
+	ldxa	[%sp + STACK_BIAS + 0x68] %asi, %i5;	\
+	ldxa	[%sp + STACK_BIAS + 0x70] %asi, %i6;	\
+	ldxa	[%sp + STACK_BIAS + 0x78] %asi, %i7;	\
+	ba,pt	%xcc, user_rtt_pre_restore;		\
+	 restored;					\
+	nop; nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop; nop;			\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;
+
+
 /* Normal 32bit fill */
 #define FILL_2_GENERIC(ASI)				\
 	srl	%sp, 0, %sp;				\
@@ -385,6 +589,34 @@
 	b,a,pt	%xcc, fill_fixup_mna;			\
 	b,a,pt	%xcc, fill_fixup;
 
+#define FILL_2_GENERIC_RTRAP				\
+user_rtt_fill_32bit:					\
+	srl	%sp, 0, %sp;				\
+	lduwa	[%sp + 0x00] %asi, %l0;			\
+	lduwa	[%sp + 0x04] %asi, %l1;			\
+	lduwa	[%sp + 0x08] %asi, %l2;			\
+	lduwa	[%sp + 0x0c] %asi, %l3;			\
+	lduwa	[%sp + 0x10] %asi, %l4;			\
+	lduwa	[%sp + 0x14] %asi, %l5;			\
+	lduwa	[%sp + 0x18] %asi, %l6;			\
+	lduwa	[%sp + 0x1c] %asi, %l7;			\
+	lduwa	[%sp + 0x20] %asi, %i0;			\
+	lduwa	[%sp + 0x24] %asi, %i1;			\
+	lduwa	[%sp + 0x28] %asi, %i2;			\
+	lduwa	[%sp + 0x2c] %asi, %i3;			\
+	lduwa	[%sp + 0x30] %asi, %i4;			\
+	lduwa	[%sp + 0x34] %asi, %i5;			\
+	lduwa	[%sp + 0x38] %asi, %i6;			\
+	lduwa	[%sp + 0x3c] %asi, %i7;			\
+	ba,pt	%xcc, user_rtt_pre_restore;		\
+	 restored;					\
+	nop; nop; nop; nop; nop;			\
+	nop; nop; nop; nop; nop;			\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;		\
+	ba,a,pt	%xcc, user_rtt_fill_fixup;
+		
+
 #define FILL_1_NORMAL FILL_1_GENERIC(ASI_AIUP)
 #define FILL_2_NORMAL FILL_2_GENERIC(ASI_AIUP)
 #define FILL_3_NORMAL FILL_0_NORMAL