ARC: entry: use gp to cache task pointer (vs. r25)

The motivation is eventual ABI considerations for ARCv3 but even without
it this change us worthwhile as diffstat reduces 100 net lines

r25 is a callee saved register, normally not saved by entry code in
pt_regs. However because of its usage in CONFIG_ARC_CURR_IN_REG it needs
to be. This in turn requires a whole bunch of special casing when we
need to access r25. Then there is distinction between user mode r25 vs.
kernel mode r25 - hence distinct SAVE_CALLEE_SAVED_{USER,KERNEL}

Instead use gp which is a scratch register and thus saved already in entry
code. This cleans things up significantly and much nocer on eyes:

 - SAVE_CALLEE_SAVED_{USER,KERNEL} are now exactly same
 - no special user_r25 slot in pt_reggs

Note that typical global asm registers are callee-saved (r25), but gp is
not callee-saved thus needs additional -ffixed-<reg> toggle

Signed-off-by: Vineet Gupta <vgupta@kernel.org>

Author: vineetgarc

arch/arc/Kconfig

@@ -492,11 +492,11 @@ config ARC_KVADDR_SIZE
 	  kernel-user gutter)
 
 config ARC_CURR_IN_REG
-	bool "Dedicate Register r25 for current_task pointer"
+	bool "cache current task pointer in gp"
 	default y
 	help
-	  This reserved Register R25 to point to Current Task in
-	  kernel mode. This saves memory access for each such access
+	  This reserves gp register to point to Current Task in
+	  kernel mode eliding memory access for each access
 
 
 config ARC_EMUL_UNALIGNED

arch/arc/Makefile

@@ -28,14 +28,14 @@ cflags-y				+= $(tune-mcpu-def-y)
 endif
 endif
 
-
 ifdef CONFIG_ARC_CURR_IN_REG
 # For a global register definition, make sure it gets passed to every file
 # We had a customer reported bug where some code built in kernel was NOT using
-# any kernel headers, and missing the r25 global register
+# any kernel headers, and missing the global register
 # Can't do unconditionally because of recursive include issues
 # due to <linux/thread_info.h>
 LINUXINCLUDE	+=  -include $(srctree)/arch/arc/include/asm/current.h
+cflags-y	+= -ffixed-gp
 endif
 
 cflags-y				+= -fsection-anchors
@@ -67,7 +67,7 @@ cflags-$(CONFIG_ARC_DW2_UNWIND)		+= -fasynchronous-unwind-tables $(cfi)
 # small data is default for elf32 tool-chain. If not usable, disable it
 # This also allows repurposing GP as scratch reg to gcc reg allocator
 disable_small_data := y
-cflags-$(disable_small_data)		+= -mno-sdata -fcall-used-gp
+cflags-$(disable_small_data)		+= -mno-sdata
 
 cflags-$(CONFIG_CPU_BIG_ENDIAN)		+= -mbig-endian
 ldflags-$(CONFIG_CPU_BIG_ENDIAN)	+= -EB

arch/arc/include/asm/current.h

@@ -13,7 +13,7 @@
 
 #ifdef CONFIG_ARC_CURR_IN_REG
 
-register struct task_struct *curr_arc asm("r25");
+register struct task_struct *curr_arc asm("gp");
 #define current (curr_arc)
 
 #else

arch/arc/include/asm/entry-arcv2.h

@@ -18,7 +18,6 @@
  *              |      orig_r0      |
  *              |      event/ECR    |
  *              |      bta          |
- *              |      user_r25     |
  *              |      gp           |
  *              |      fp           |
  *              |      sp           |
@@ -56,7 +55,7 @@
 	;                 hardware does even if CONFIG_ARC_IRQ_NO_AUTOSAVE
 	;   4. Auto save: (optional) r0-r11, blink, LPE,LPS,LPC, JLI,LDI,EI
 	;
-	; (B) Manually saved some regs: r12,r25,r30, sp,fp,gp, ACCL pair
+	; (B) Manually saved some regs: r12,r30, sp,fp,gp, ACCL pair
 
 #ifdef CONFIG_ARC_IRQ_NO_AUTOSAVE
 	; carve pt_regs on stack (case #3), PC/STAT32 already on stack
@@ -157,17 +156,17 @@
 
 	st	r10, [sp, PT_sp]	; SP (pt_regs->sp)
 
-#ifdef CONFIG_ARC_CURR_IN_REG
-	st	r25, [sp, PT_user_r25]
-	GET_CURR_TASK_ON_CPU	r25
-#endif
-
 #ifdef CONFIG_ARC_HAS_ACCL_REGS
 	ST2	r58, r59, PT_r58
 #endif
 
 	/* clobbers r10, r11 registers pair */
 	DSP_SAVE_REGFILE_IRQ
+
+#ifdef CONFIG_ARC_CURR_IN_REG
+	GET_CURR_TASK_ON_CPU	gp
+#endif
+
 .endm
 
 /*------------------------------------------------------------------------*/
@@ -188,10 +187,6 @@
 	sr	r10, [AUX_USER_SP]
 1:
 
-#ifdef CONFIG_ARC_CURR_IN_REG
-	ld	r25, [sp, PT_user_r25]
-#endif
-
 	/* clobbers r10, r11 registers pair */
 	DSP_RESTORE_REGFILE_IRQ
 

arch/arc/include/asm/entry-compact.h

@@ -151,14 +151,6 @@
 	/* ARC700 doesn't provide auto-stack switching */
 	SWITCH_TO_KERNEL_STK
 
-#ifdef CONFIG_ARC_CURR_IN_REG
-	/* Treat r25 as scratch reg (save on stack) and load with "current" */
-	PUSH    r25
-	GET_CURR_TASK_ON_CPU   r25
-#else
-	sub     sp, sp, 4
-#endif
-
 	st.a	r0, [sp, -8]    /* orig_r0 needed for syscall (skip ECR slot) */
 	sub	sp, sp, 4	/* skip pt_regs->sp, already saved above */
 
@@ -179,6 +171,11 @@
 
 	lr	r10, [ecr]
 	st      r10, [sp, PT_event]    /* EV_Trap expects r10 to have ECR */
+
+#ifdef CONFIG_ARC_CURR_IN_REG
+	/* gp already saved on stack: now load with "current" */
+	GET_CURR_TASK_ON_CPU   gp
+#endif
 .endm
 
 /*--------------------------------------------------------------
@@ -208,11 +205,8 @@
 	POP	gp
 	RESTORE_R12_TO_R0
 
-#ifdef CONFIG_ARC_CURR_IN_REG
-	ld	r25, [sp, 12]
-#endif
 	ld  sp, [sp] /* restore original sp */
-	/* orig_r0, ECR, user_r25 skipped automatically */
+	/* orig_r0, ECR skipped automatically */
 .endm
 
 /* Dummy ECR values for Interrupts */
@@ -229,13 +223,6 @@
 
 	SWITCH_TO_KERNEL_STK
 
-#ifdef CONFIG_ARC_CURR_IN_REG
-	/* Treat r25 as scratch reg (save on stack) and load with "current" */
-	PUSH    r25
-	GET_CURR_TASK_ON_CPU   r25
-#else
-	sub     sp, sp, 4
-#endif
 
 	PUSH	0x003\LVL\()abcd    /* Dummy ECR */
 	sub	sp, sp, 8	    /* skip orig_r0 (not needed)
@@ -255,6 +242,10 @@
 	PUSHAX	lp_start
 	PUSHAX	bta_l\LVL\()
 
+#ifdef CONFIG_ARC_CURR_IN_REG
+	/* gp already saved on stack: now load with "current" */
+	GET_CURR_TASK_ON_CPU   gp
+#endif
 .endm
 
 /*--------------------------------------------------------------
@@ -282,11 +273,7 @@
 	POP	gp
 	RESTORE_R12_TO_R0
 
-#ifdef CONFIG_ARC_CURR_IN_REG
-	ld	r25, [sp, 12]
-#endif
-	ld  sp, [sp] /* restore original sp */
-	/* orig_r0, ECR, user_r25 skipped automatically */
+	ld  sp, [sp] /* restore original sp; orig_r0, ECR skipped implicitly */
 .endm
 
 /* Get thread_info of "current" tsk */

arch/arc/include/asm/entry.h

@@ -91,7 +91,7 @@
  * Helpers to save/restore callee-saved regs:
  * used by several macros below
  *-------------------------------------------------------------*/
-.macro SAVE_R13_TO_R24
+.macro SAVE_R13_TO_R25
 	PUSH	r13
 	PUSH	r14
 	PUSH	r15
@@ -104,9 +104,11 @@
 	PUSH	r22
 	PUSH	r23
 	PUSH	r24
+	PUSH	r25
 .endm
 
-.macro RESTORE_R24_TO_R13
+.macro RESTORE_R25_TO_R13
+	POP	r25
 	POP	r24
 	POP	r23
 	POP	r22
@@ -121,81 +123,31 @@
 	POP	r13
 .endm
 
-/*--------------------------------------------------------------
- * Collect User Mode callee regs as struct callee_regs - needed by
- * fork/do_signal/unaligned-access-emulation.
- * (By default only scratch regs are saved on entry to kernel)
- *
- * Special handling for r25 if used for caching Task Pointer.
- * It would have been saved in task->thread.user_r25 already, but to keep
- * the interface same it is copied into regular r25 placeholder in
- * struct callee_regs.
- *-------------------------------------------------------------*/
+/*
+ * save user mode callee regs as struct callee_regs
+ *  - needed by fork/do_signal/unaligned-access-emulation.
+ */
 .macro SAVE_CALLEE_SAVED_USER
+	SAVE_R13_TO_R25
+.endm
 
-	mov	r12, sp		; save SP as ref to pt_regs
-	SAVE_R13_TO_R24
-
-#ifdef CONFIG_ARC_CURR_IN_REG
-	; Retrieve orig r25 and save it with rest of callee_regs
-	ld	r12, [r12, PT_user_r25]
-	PUSH	r12
-#else
-	PUSH	r25
-#endif
-
+/*
+ * restore user mode callee regs as struct callee_regs
+ *  - could have been changed by ptrace tracer or unaligned-access fixup
+ */
+.macro RESTORE_CALLEE_SAVED_USER
+	RESTORE_R25_TO_R13
 .endm
 
-/*--------------------------------------------------------------
- * Save kernel Mode callee regs at the time of Contect Switch.
- *
- * Special handling for r25 if used for caching Task Pointer.
- * Kernel simply skips saving it since it will be loaded with
- * incoming task pointer anyways
- *-------------------------------------------------------------*/
+/*
+ * save/restore kernel mode callee regs at the time of context switch
+ */
 .macro SAVE_CALLEE_SAVED_KERNEL
-
-	SAVE_R13_TO_R24
-
-#ifdef CONFIG_ARC_CURR_IN_REG
-	sub     sp, sp, 4
-#else
-	PUSH	r25
-#endif
+	SAVE_R13_TO_R25
 .endm
 
-/*--------------------------------------------------------------
- * Opposite of SAVE_CALLEE_SAVED_KERNEL
- *-------------------------------------------------------------*/
 .macro RESTORE_CALLEE_SAVED_KERNEL
-
-#ifdef CONFIG_ARC_CURR_IN_REG
-	add     sp, sp, 4  /* skip usual r25 placeholder */
-#else
-	POP	r25
-#endif
-	RESTORE_R24_TO_R13
-.endm
-
-/*--------------------------------------------------------------
- * Opposite of SAVE_CALLEE_SAVED_USER
- *
- * ptrace tracer or unaligned-access fixup might have changed a user mode
- * callee reg which is saved back to usual r25 storage location
- *-------------------------------------------------------------*/
-.macro RESTORE_CALLEE_SAVED_USER
-
-#ifdef CONFIG_ARC_CURR_IN_REG
-	POP	r12
-#else
-	POP	r25
-#endif
-	RESTORE_R24_TO_R13
-
-	; SP is back to start of pt_regs
-#ifdef CONFIG_ARC_CURR_IN_REG
-	st	r12, [sp, PT_user_r25]
-#endif
+	RESTORE_R25_TO_R13
 .endm
 
 /*--------------------------------------------------------------
@@ -231,10 +183,10 @@
 
 #ifdef CONFIG_SMP
 
-/*-------------------------------------------------
+/*
  * Retrieve the current running task on this CPU
- * 1. Determine curr CPU id.
- * 2. Use it to index into _current_task[ ]
+ *  - loads it from backing _current_task[] (and can't use the
+ *    caching reg for current task
  */
 .macro  GET_CURR_TASK_ON_CPU   reg
 	GET_CPU_ID  \reg
@@ -256,7 +208,7 @@
 	add2 \tmp, @_current_task, \tmp
 	st   \tsk, [\tmp]
 #ifdef CONFIG_ARC_CURR_IN_REG
-	mov r25, \tsk
+	mov gp, \tsk
 #endif
 
 .endm
@@ -271,21 +223,20 @@
 .macro  SET_CURR_TASK_ON_CPU    tsk, tmp
 	st  \tsk, [@_current_task]
 #ifdef CONFIG_ARC_CURR_IN_REG
-	mov r25, \tsk
+	mov gp, \tsk
 #endif
 .endm
 
 #endif /* SMP / UNI */
 
-/* ------------------------------------------------------------------
+/*
  * Get the ptr to some field of Current Task at @off in task struct
- *  -Uses r25 for Current task ptr if that is enabled
+ *  - Uses current task cached in reg if enabled
  */
-
 #ifdef CONFIG_ARC_CURR_IN_REG
 
 .macro GET_CURR_TASK_FIELD_PTR  off,  reg
-	add \reg, r25, \off
+	add \reg, gp, \off
 .endm
 
 #else

arch/arc/include/asm/ptrace.h

@@ -52,11 +52,9 @@ struct pt_regs {
 		};
 		unsigned long event;
 	};
-
-	unsigned long user_r25;
 };
 
-#define MAX_REG_OFFSET offsetof(struct pt_regs, user_r25)
+#define MAX_REG_OFFSET offsetof(struct pt_regs, event)
 
 #else
 
@@ -79,8 +77,6 @@ struct pt_regs {
 
 	unsigned long bta;	/* bta_l1, bta_l2, erbta */
 
-	unsigned long user_r25;
-
 	unsigned long r26;	/* gp */
 	unsigned long fp;
 	unsigned long sp;	/* user/kernel sp depending on where we came from  */

arch/arc/kernel/asm-offsets.c

@@ -63,8 +63,6 @@ int main(void)
 	DEFINE(PT_blink, offsetof(struct pt_regs, blink));
 	DEFINE(PT_lpe, offsetof(struct pt_regs, lp_end));
 	DEFINE(PT_lpc, offsetof(struct pt_regs, lp_count));
-	DEFINE(PT_user_r25, offsetof(struct pt_regs, user_r25));
-
 	DEFINE(SZ_CALLEE_REGS, sizeof(struct callee_regs));
 	DEFINE(SZ_PT_REGS, sizeof(struct pt_regs));