ARCv2: entry: rewrite to enable use of double load/stores LDD/STD

 - the motivation was to be remove blatent copy-paste due to hasty support
   of CONFIG_ARC_IRQ_NO_AUTOSAVE support

 - but with refactoring we could use LDD/STD to greatly optimize the code

Signed-off-by: Vineet Gupta <vgupta@synopsys.com>

2 files changed, 158 insertions, 157 deletions

diff --git a/arch/arc/include/asm/entry-arcv2.h b/arch/arc/include/asm/entry-arcv2.h
index beaf655666cb..0733752ce7fe 100644
--- a/arch/arc/include/asm/entry-arcv2.h
+++ b/arch/arc/include/asm/entry-arcv2.h
@@ -46,7 +46,8 @@
  */
 
 /*------------------------------------------------------------------------*/
-.macro INTERRUPT_PROLOGUE	called_from
+.macro INTERRUPT_PROLOGUE
+
 	; (A) Before jumping to Interrupt Vector, hardware micro-ops did following:
 	;   1. SP auto-switched to kernel mode stack
 	;   2. STATUS32.Z flag set if in U mode at time of interrupt (U:1,K:0)
@@ -57,39 +58,87 @@
 	; (B) Manually saved some regs: r12,r25,r30, sp,fp,gp, ACCL pair
 
 #ifdef CONFIG_ARC_IRQ_NO_AUTOSAVE
-.ifnc \called_from, exception
-	st.as	r9, [sp, -10]	; save r9 in it's final stack slot
-	sub	sp, sp, 12	; skip JLI, LDI, EI
-
-	PUSH	lp_count
-	PUSHAX	lp_start
-	PUSHAX	lp_end
-	PUSH	blink
-
-	PUSH	r11
-	PUSH	r10
-
-	sub	sp, sp, 4	; skip r9
-
-	PUSH	r8
-	PUSH	r7
-	PUSH	r6
-	PUSH	r5
-	PUSH	r4
-	PUSH	r3
-	PUSH	r2
-	PUSH	r1
-	PUSH	r0
-.endif
-#endif
+	; carve pt_regs on stack (case #3), PC/STAT32 already on stack
+	sub	sp, sp, SZ_PT_REGS - 8
 
-#ifdef CONFIG_ARC_HAS_ACCL_REGS
-	PUSH	r59
-	PUSH	r58
+	__SAVE_REGFILE_HARD
+#else
+	; carve pt_regs on stack (case #4), which grew partially already
+	sub	sp, sp, PT_r0
 #endif
 
-	PUSH	r30
-	PUSH	r12
+	__SAVE_REGFILE_SOFT
+.endm
+
+/*------------------------------------------------------------------------*/
+.macro EXCEPTION_PROLOGUE
+
+	; (A) Before jumping to Exception Vector, hardware micro-ops did following:
+	;   1. SP auto-switched to kernel mode stack
+	;   2. STATUS32.Z flag set if in U mode at time of exception (U:1,K:0)
+	;
+	; (B) Manually save the complete reg file below
+
+	sub	sp, sp, SZ_PT_REGS	; carve pt_regs
+
+	; _HARD saves r10 clobbered by _SOFT as scratch hence comes first
+
+	__SAVE_REGFILE_HARD
+	__SAVE_REGFILE_SOFT
+
+	st	r0, [sp]	; orig_r0
+
+	lr	r10, [eret]
+	lr	r11, [erstatus]
+	ST2	r10, r11, PT_ret
+
+	lr	r10, [ecr]
+	lr	r11, [erbta]
+	ST2	r10, r11, PT_event
+	mov	r9, r10
+
+	; OUTPUT: r9 has ECR
+.endm
+
+/*------------------------------------------------------------------------
+ * This macro saves the registers manually which would normally be autosaved
+ * by hardware on taken interrupts. It is used by
+ *   - exception handlers (which don't have autosave)
+ *   - interrupt autosave disabled due to CONFIG_ARC_IRQ_NO_AUTOSAVE
+ */
+.macro __SAVE_REGFILE_HARD
+
+	ST2	r0,  r1,  PT_r0
+	ST2	r2,  r3,  PT_r2
+	ST2	r4,  r5,  PT_r4
+	ST2	r6,  r7,  PT_r6
+	ST2	r8,  r9,  PT_r8
+	ST2	r10, r11, PT_r10
+
+	st	blink, [sp, PT_blink]
+
+	lr	r10, [lp_end]
+	lr	r11, [lp_start]
+	ST2	r10, r11, PT_lpe
+
+	st	lp_count, [sp, PT_lpc]
+
+	; skip JLI, LDI, EI for now
+.endm
+
+/*------------------------------------------------------------------------
+ * This macros saves a bunch of other registers which can't be autosaved for
+ * various reasons:
+ *   - r12: the last caller saved scratch reg since hardware saves in pairs so r0-r11
+ *   - r30: free reg, used by gcc as scratch
+ *   - ACCL/ACCH pair when they exist
+ */
+.macro __SAVE_REGFILE_SOFT
+
+	ST2	gp, fp, PT_r26		; gp (r26), fp (r27)
+
+	st	r12, [sp, PT_sp + 4]
+	st	r30, [sp, PT_sp + 8]
 
 	; Saving pt_regs->sp correctly requires some extra work due to the way
 	; Auto stack switch works
@@ -100,46 +149,32 @@
 	; 2. Upon entry SP is always saved (for any inspection, unwinding etc),
 	;    but on return, restored only if U mode
 
-	lr	r9, [AUX_USER_SP]			; U mode SP
+	lr	r10, [AUX_USER_SP]	; U mode SP
 
-	mov.nz	r9, sp
-	add.nz	r9, r9, SZ_PT_REGS - PT_sp - 4		; K mode SP
+	; ISA requires ADD.nz to have same dest and src reg operands
+	mov.nz	r10, sp
+	add.nz	r10, r10, SZ_PT_REGS	; K mode SP
 
-	PUSH	r9					; SP (pt_regs->sp)
-
-	PUSH	fp
-	PUSH	gp
+	st	r10, [sp, PT_sp]	; SP (pt_regs->sp)
 
 #ifdef CONFIG_ARC_CURR_IN_REG
-	PUSH	r25			; user_r25
+	st	r25, [sp, PT_user_r25]
 	GET_CURR_TASK_ON_CPU	r25
-#else
-	sub	sp, sp, 4
 #endif
 
-.ifnc \called_from, exception
-	sub	sp, sp, 12	; BTA/ECR/orig_r0 placeholder per pt_regs
-.endif
+#ifdef CONFIG_ARC_HAS_ACCL_REGS
+	ST2	r58, r59, PT_sp + 12
+#endif
 
 .endm
 
 /*------------------------------------------------------------------------*/
-.macro INTERRUPT_EPILOGUE	called_from
+.macro __RESTORE_REGFILE_SOFT
 
-	; INPUT: r0 has STAT32 of calling context
-	; INPUT: Z flag set if returning to K mode
-.ifnc \called_from, exception
-	add	sp, sp, 12	; skip BTA/ECR/orig_r0 placeholderss
-.endif
-
-#ifdef CONFIG_ARC_CURR_IN_REG
-	POP	r25
-#else
-	add	sp, sp, 4
-#endif
+	LD2	gp, fp, PT_r26		; gp (r26), fp (r27)
 
-	POP	gp
-	POP	fp
+	ld	r12, [sp, PT_sp + 4]
+	ld	r30, [sp, PT_sp + 8]
 
 	; Restore SP (into AUX_USER_SP) only if returning to U mode
 	;  - for K mode, it will be implicitly restored as stack is unwound
@@ -147,129 +182,77 @@
 	;    but that doesn't really matter
 	bz	1f
 
-	POPAX	AUX_USER_SP
+	ld	r10, [sp, PT_sp]	; SP (pt_regs->sp)
+	sr	r10, [AUX_USER_SP]
 1:
-	POP	r12
-	POP	r30
 
-#ifdef CONFIG_ARC_HAS_ACCL_REGS
-	POP	r58
-	POP	r59
+#ifdef CONFIG_ARC_CURR_IN_REG
+	ld	r25, [sp, PT_user_r25]
 #endif
 
-#ifdef CONFIG_ARC_IRQ_NO_AUTOSAVE
-.ifnc \called_from, exception
-	POP	r0
-	POP	r1
-	POP	r2
-	POP	r3
-	POP	r4
-	POP	r5
-	POP	r6
-	POP	r7
-	POP	r8
-	POP	r9
-	POP	r10
-	POP	r11
-
-	POP	blink
-	POPAX	lp_end
-	POPAX	lp_start
-
-	POP	r9
-	mov	lp_count, r9
-
-	add	sp, sp, 12	; skip JLI, LDI, EI
-	ld.as	r9, [sp, -10]	; reload r9 which got clobbered
-.endif
+#ifdef CONFIG_ARC_HAS_ACCL_REGS
+	LD2	r58, r59, PT_sp + 12
 #endif
-
 .endm
 
 /*------------------------------------------------------------------------*/
-.macro EXCEPTION_PROLOGUE
+.macro __RESTORE_REGFILE_HARD
 
-	; (A) Before jumping to Exception Vector, hardware micro-ops did following:
-	;   1. SP auto-switched to kernel mode stack
-	;   2. STATUS32.Z flag set if in U mode at time of exception (U:1,K:0)
-	;
-	; (B) Manually save the complete reg file below
+	ld	blink, [sp, PT_blink]
 
-	PUSH	r9		; freeup a register: slot of erstatus
+	LD2	r10, r11, PT_lpe
+	sr	r10, [lp_end]
+	sr	r11, [lp_start]
 
-	PUSHAX	eret
-	sub	sp, sp, 12	; skip JLI, LDI, EI
-	PUSH	lp_count
-	PUSHAX	lp_start
-	PUSHAX	lp_end
-	PUSH	blink
+	ld	r10, [sp, PT_lpc]	; lp_count can't be target of LD
+	mov	lp_count, r10
 
-	PUSH	r11
-	PUSH	r10
+	LD2	r0,  r1,  PT_r0
+	LD2	r2,  r3,  PT_r2
+	LD2	r4,  r5,  PT_r4
+	LD2	r6,  r7,  PT_r6
+	LD2	r8,  r9,  PT_r8
+	LD2	r10, r11, PT_r10
+.endm
 
-	ld.as	r9,  [sp, 10]	; load stashed r9 (status32 stack slot)
-	lr	r10, [erstatus]
-	st.as	r10, [sp, 10]	; save status32 at it's right stack slot
 
-	PUSH	r9
-	PUSH	r8
-	PUSH	r7
-	PUSH	r6
-	PUSH	r5
-	PUSH	r4
-	PUSH	r3
-	PUSH	r2
-	PUSH	r1
-	PUSH	r0
+/*------------------------------------------------------------------------*/
+.macro INTERRUPT_EPILOGUE
 
-	; -- for interrupts, regs above are auto-saved by h/w in that order --
-	; Now do what ISR prologue does (manually save r12, sp, fp, gp, r25)
+	; INPUT: r0 has STAT32 of calling context
+	; INPUT: Z flag set if returning to K mode
 
-	INTERRUPT_PROLOGUE  exception
+	; _SOFT clobbers r10 restored by _HARD hence the order
 
-	PUSHAX	erbta
-	PUSHAX	ecr		; r9 contains ECR, expected by EV_Trap
+	__RESTORE_REGFILE_SOFT
+
+#ifdef CONFIG_ARC_IRQ_NO_AUTOSAVE
+	__RESTORE_REGFILE_HARD
+	add	sp, sp, SZ_PT_REGS - 8
+#else
+	add	sp, sp, PT_r0
+#endif
 
-	PUSH	r0		; orig_r0
-	; OUTPUT: r9 has ECR
 .endm
 
 /*------------------------------------------------------------------------*/
 .macro EXCEPTION_EPILOGUE
 
 	; INPUT: r0 has STAT32 of calling context
-	btst   r0, STATUS_U_BIT	; Z flag set if K, used in INTERRUPT_EPILOGUE
-
-	add	sp, sp, 8	; orig_r0/ECR don't need restoring
-	POPAX	erbta
-
-	INTERRUPT_EPILOGUE  exception
-
-	POP	r0
-	POP	r1
-	POP	r2
-	POP	r3
-	POP	r4
-	POP	r5
-	POP	r6
-	POP	r7
-	POP	r8
-	POP	r9
-	POP	r10
-	POP	r11
-
-	POP	blink
-	POPAX	lp_end
-	POPAX	lp_start
-
-	POP	r9
-	mov	lp_count, r9
-
-	add	sp, sp, 12	; skip JLI, LDI, EI
-	POPAX	eret
-	POPAX	erstatus
-
-	ld.as	r9, [sp, -12]	; reload r9 which got clobbered
+
+	btst	r0, STATUS_U_BIT	; Z flag set if K, used in restoring SP
+
+	ld	r10, [sp, PT_event + 4]
+	sr	r10, [erbta]
+
+	LD2	r10, r11, PT_ret
+	sr	r10, [eret]
+	sr	r11, [erstatus]
+
+	__RESTORE_REGFILE_SOFT
+	__RESTORE_REGFILE_HARD
+
+	add	sp, sp, SZ_PT_REGS
 .endm
 
 .macro FAKE_RET_FROM_EXCPN
diff --git a/arch/arc/include/asm/linkage.h b/arch/arc/include/asm/linkage.h
index 54f5ec5c1759..a0eeb9f8f0a9 100644
--- a/arch/arc/include/asm/linkage.h
+++ b/arch/arc/include/asm/linkage.h
@@ -10,6 +10,24 @@
 
 #ifdef __ASSEMBLY__
 
+.macro ST2 e, o, off
+#ifdef CONFIG_ARC_HAS_LL64
+	std	\e, [sp, \off]
+#else
+	st	\e, [sp, \off]
+	st	\o, [sp, \off+4]
+#endif
+.endm
+
+.macro LD2 e, o, off
+#ifdef CONFIG_ARC_HAS_LL64
+	ldd	\e, [sp, \off]
+#else
+	ld	\e, [sp, \off]
+	ld	\o, [sp, \off+4]
+#endif
+.endm
+
 #define ASM_NL		 `	/* use '`' to mark new line in macro */
 
 /* annotation for data we want in DCCM - if enabled in .config */