ARC: Process-creation/scheduling/idle-loop

Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>

arch/arc/Kconfig

@@ -17,6 +17,8 @@ config ARC
 	select GENERIC_FIND_FIRST_BIT
 	# for now, we don't need GENERIC_IRQ_PROBE, CONFIG_GENERIC_IRQ_CHIP
 	select GENERIC_IRQ_SHOW
+	select GENERIC_KERNEL_EXECVE
+	select GENERIC_KERNEL_THREAD
 	select GENERIC_PENDING_IRQ if SMP
 	select GENERIC_SMP_IDLE_THREAD
 	select HAVE_GENERIC_HARDIRQS

arch/arc/include/asm/arcregs.h

@@ -47,6 +47,17 @@
 #define AUX_ITRIGGER		0x40d
 #define AUX_IPULSE		0x415
 
+/*
+ * Floating Pt Registers
+ * Status regs are read-only (build-time) so need not be saved/restored
+ */
+#define ARC_AUX_FP_STAT         0x300
+#define ARC_AUX_DPFP_1L         0x301
+#define ARC_AUX_DPFP_1H         0x302
+#define ARC_AUX_DPFP_2L         0x303
+#define ARC_AUX_DPFP_2H         0x304
+#define ARC_AUX_DPFP_STAT       0x305
+
 #ifndef __ASSEMBLY__
 
 /*
@@ -110,6 +121,15 @@
 
 #endif
 
+#ifdef CONFIG_ARC_FPU_SAVE_RESTORE
+/* These DPFP regs need to be saved/restored across ctx-sw */
+struct arc_fpu {
+	struct {
+		unsigned int l, h;
+	} aux_dpfp[2];
+};
+#endif
+
 #endif /* __ASEMBLY__ */
 
 #endif /* __KERNEL__ */

arch/arc/include/asm/processor.h

@@ -29,6 +29,9 @@ struct thread_struct {
 	unsigned long callee_reg;	/* pointer to callee regs */
 	unsigned long fault_address;	/* dbls as brkpt holder as well */
 	unsigned long cause_code;	/* Exception Cause Code (ECR) */
+#ifdef CONFIG_ARC_FPU_SAVE_RESTORE
+	struct arc_fpu fpu;
+#endif
 };
 
 #define INIT_THREAD  {                          \
@@ -54,12 +57,6 @@ unsigned long thread_saved_pc(struct task_struct *t);
 
 #define cpu_relax()	do { } while (0)
 
-/*
- * Create a new kernel thread
- */
-
-extern int kernel_thread(int (*fn) (void *), void *arg, unsigned long flags);
-
 #define copy_segments(tsk, mm)      do { } while (0)
 #define release_segments(mm)        do { } while (0)
 

arch/arc/include/asm/ptrace.h

@@ -91,6 +91,14 @@ struct callee_regs {
 #define in_syscall(regs) (((regs->orig_r8) >= 0 && \
 			   (regs->orig_r8 <= NR_syscalls)) ? 1 : 0)
 
+#define current_pt_regs()					\
+({								\
+	/* open-coded current_thread_info() */			\
+	register unsigned long sp asm ("sp");			\
+	unsigned long pg_start = (sp & ~(THREAD_SIZE - 1));	\
+	(struct pt_regs *)(pg_start + THREAD_SIZE - 4) - 1;	\
+})
+
 #endif /* !__ASSEMBLY__ */
 
 #endif /* __KERNEL__ */

arch/arc/include/asm/switch_to.h

@@ -0,0 +1,41 @@
+/*
+ * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef _ASM_ARC_SWITCH_TO_H
+#define _ASM_ARC_SWITCH_TO_H
+
+#ifndef __ASSEMBLY__
+
+#include <linux/sched.h>
+
+#ifdef CONFIG_ARC_FPU_SAVE_RESTORE
+
+extern void fpu_save_restore(struct task_struct *p, struct task_struct *n);
+#define ARC_FPU_PREV(p, n)	fpu_save_restore(p, n)
+#define ARC_FPU_NEXT(t)
+
+#else
+
+#define ARC_FPU_PREV(p, n)
+#define ARC_FPU_NEXT(n)
+
+#endif /* !CONFIG_ARC_FPU_SAVE_RESTORE */
+
+struct task_struct *__switch_to(struct task_struct *p, struct task_struct *n);
+
+#define switch_to(prev, next, last)	\
+do {					\
+	ARC_FPU_PREV(prev, next);	\
+	last = __switch_to(prev, next);\
+	ARC_FPU_NEXT(next);		\
+	mb();				\
+} while (0)
+
+#endif
+
+#endif

arch/arc/kernel/ctx_sw.c

@@ -0,0 +1,91 @@
+/*
+ * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Vineetg: Aug 2009
+ *  -"C" version of lowest level context switch asm macro called by schedular
+ *   gcc doesn't generate the dward CFI info for hand written asm, hence can't
+ *   backtrace out of it (e.g. tasks sleeping in kernel).
+ *   So we cheat a bit by writing almost similar code in inline-asm.
+ *  -This is a hacky way of doing things, but there is no other simple way.
+ *   I don't want/intend to extend unwinding code to understand raw asm
+ */
+
+#include <asm/asm-offsets.h>
+#include <linux/sched.h>
+
+struct task_struct *__sched
+__switch_to(struct task_struct *prev_task, struct task_struct *next_task)
+{
+	unsigned int tmp;
+	unsigned int prev = (unsigned int)prev_task;
+	unsigned int next = (unsigned int)next_task;
+	int num_words_to_skip = 1;
+
+	__asm__ __volatile__(
+		/* FP/BLINK save generated by gcc (standard function prologue */
+		"st.a    r13, [sp, -4]   \n\t"
+		"st.a    r14, [sp, -4]   \n\t"
+		"st.a    r15, [sp, -4]   \n\t"
+		"st.a    r16, [sp, -4]   \n\t"
+		"st.a    r17, [sp, -4]   \n\t"
+		"st.a    r18, [sp, -4]   \n\t"
+		"st.a    r19, [sp, -4]   \n\t"
+		"st.a    r20, [sp, -4]   \n\t"
+		"st.a    r21, [sp, -4]   \n\t"
+		"st.a    r22, [sp, -4]   \n\t"
+		"st.a    r23, [sp, -4]   \n\t"
+		"st.a    r24, [sp, -4]   \n\t"
+		"st.a    r25, [sp, -4]   \n\t"
+		"sub     sp, sp, %4      \n\t"	/* create gutter at top */
+
+		/* set ksp of outgoing task in tsk->thread.ksp */
+		"st.as   sp, [%3, %1]    \n\t"
+
+		"sync   \n\t"
+
+		/*
+		 * setup _current_task with incoming tsk.
+		 * optionally, set r25 to that as well
+		 * For SMP extra work to get to &_current_task[cpu]
+		 * (open coded SET_CURR_TASK_ON_CPU)
+		 */
+		"st  %2, [@_current_task]	\n\t"
+
+		/* get ksp of incoming task from tsk->thread.ksp */
+		"ld.as  sp, [%2, %1]   \n\t"
+
+		/* start loading it's CALLEE reg file */
+
+		"add    sp, sp, %4     \n\t"	/* skip gutter at top */
+
+		"ld.ab   r25, [sp, 4]   \n\t"
+		"ld.ab   r24, [sp, 4]   \n\t"
+		"ld.ab   r23, [sp, 4]   \n\t"
+		"ld.ab   r22, [sp, 4]   \n\t"
+		"ld.ab   r21, [sp, 4]   \n\t"
+		"ld.ab   r20, [sp, 4]   \n\t"
+		"ld.ab   r19, [sp, 4]   \n\t"
+		"ld.ab   r18, [sp, 4]   \n\t"
+		"ld.ab   r17, [sp, 4]   \n\t"
+		"ld.ab   r16, [sp, 4]   \n\t"
+		"ld.ab   r15, [sp, 4]   \n\t"
+		"ld.ab   r14, [sp, 4]   \n\t"
+		"ld.ab   r13, [sp, 4]   \n\t"
+
+		/* last (ret value) = prev : although for ARC it mov r0, r0 */
+		"mov     %0, %3        \n\t"
+
+		/* FP/BLINK restore generated by gcc (standard func epilogue */
+
+		: "=r"(tmp)
+		: "n"((TASK_THREAD + THREAD_KSP) / 4), "r"(next), "r"(prev),
+		  "n"(num_words_to_skip * 4)
+		: "blink"
+	);
+
+	return (struct task_struct *)tmp;
+}

arch/arc/kernel/ctx_sw_asm.S

@@ -0,0 +1,58 @@
+/*
+ * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Vineetg: Aug 2009
+ *  -Moved core context switch macro out of entry.S into this file.
+ *  -This is the more "natural" hand written assembler
+ */
+
+#include <asm/entry.h>       /* For the SAVE_* macros */
+#include <asm/asm-offsets.h>
+#include <asm/linkage.h>
+
+;################### Low Level Context Switch ##########################
+
+	.section .sched.text,"ax",@progbits
+	.align 4
+	.global __switch_to
+	.type   __switch_to, @function
+__switch_to:
+
+	/* Save regs on kernel mode stack of task */
+	st.a    blink, [sp, -4]
+	st.a    fp, [sp, -4]
+	SAVE_CALLEE_SAVED_KERNEL
+
+	/* Save the now KSP in task->thread.ksp */
+	st.as  sp, [r0, (TASK_THREAD + THREAD_KSP)/4]
+
+	/*
+	* Return last task in r0 (return reg)
+	* On ARC, Return reg = First Arg reg = r0.
+	* Since we already have last task in r0,
+	* don't need to do anything special to return it
+	*/
+
+	/* hardware memory barrier */
+	sync
+
+	/*
+	 * switch to new task, contained in r1
+	 * Temp reg r3 is required to get the ptr to store val
+	 */
+	SET_CURR_TASK_ON_CPU  r1, r3
+
+	/* reload SP with kernel mode stack pointer in task->thread.ksp */
+	ld.as  sp, [r1, (TASK_THREAD + THREAD_KSP)/4]
+
+	/* restore the registers */
+	RESTORE_CALLEE_SAVED_KERNEL
+	ld.ab   fp, [sp, 4]
+	ld.ab   blink, [sp, 4]
+	j       [blink]
+
+ARC_EXIT __switch_to

arch/arc/kernel/entry.S

@@ -566,8 +566,19 @@ ARC_ENTRY ret_from_fork
 	; when the forked child comes here from the __switch_to function
 	; r0 has the last task pointer.
 	; put last task in scheduler queue
-	bl  @schedule_tail
-	b @ret_from_exception
+	bl   @schedule_tail
+
+	; If kernel thread, jump to it's entry-point
+	ld   r9, [sp, PT_status32]
+	brne r9, 0, 1f
+
+	jl.d [r14]
+	mov  r0, r13		; arg to payload
+
+1:
+	; special case of kernel_thread entry point returning back due to
+	; kernel_execve() - pretend return from syscall to ret to userland
+	b    ret_from_exception
 ARC_EXIT ret_from_fork
 
 ;################### Special Sys Call Wrappers ##########################

arch/arc/kernel/fpu.c

@@ -0,0 +1,55 @@
+/*
+ * fpu.c - save/restore of Floating Point Unit Registers on task switch
+ *
+ * Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/sched.h>
+#include <asm/switch_to.h>
+
+/*
+ * To save/restore FPU regs, simplest scheme would use LR/SR insns.
+ * However since SR serializes the pipeline, an alternate "hack" can be used
+ * which uses the FPU Exchange insn (DEXCL) to r/w FPU regs.
+ *
+ * Store to 64bit dpfp1 reg from a pair of core regs:
+ *   dexcl1 0, r1, r0  ; where r1:r0 is the 64 bit val
+ *
+ * Read from dpfp1 into pair of core regs (w/o clobbering dpfp1)
+ *   mov_s    r3, 0
+ *   daddh11  r1, r3, r3   ; get "hi" into r1 (dpfp1 unchanged)
+ *   dexcl1   r0, r1, r3   ; get "low" into r0 (dpfp1 low clobbered)
+ *   dexcl1    0, r1, r0   ; restore dpfp1 to orig value
+ *
+ * However we can tweak the read, so that read-out of outgoing task's FPU regs
+ * and write of incoming task's regs happen in one shot. So all the work is
+ * done before context switch
+ */
+
+void fpu_save_restore(struct task_struct *prev, struct task_struct *next)
+{
+	unsigned int *saveto = &prev->thread.fpu.aux_dpfp[0].l;
+	unsigned int *readfrom = &next->thread.fpu.aux_dpfp[0].l;
+
+	const unsigned int zero = 0;
+
+	__asm__ __volatile__(
+		"daddh11  %0, %2, %2\n"
+		"dexcl1   %1, %3, %4\n"
+		: "=&r" (*(saveto + 1)), /* early clobber must here */
+		  "=&r" (*(saveto))
+		: "r" (zero), "r" (*(readfrom + 1)), "r" (*(readfrom))
+	);
+
+	__asm__ __volatile__(
+		"daddh22  %0, %2, %2\n"
+		"dexcl2   %1, %3, %4\n"
+		: "=&r"(*(saveto + 3)),	/* early clobber must here */
+		  "=&r"(*(saveto + 2))
+		: "r" (zero), "r" (*(readfrom + 3)), "r" (*(readfrom + 2))
+	);
+}