ARC: SMP support

ARC common code to enable a SMP system + ISS provided SMP extensions.

ARC700 natively lacks SMP support, hence some of the core features are
are only enabled if SoCs have the necessary h/w pixie-dust. This
includes:
-Inter Processor Interrupts (IPI)
-Cache coherency
-load-locked/store-conditional
...

The low level exception handling would be completely broken in SMP
because we don't have hardware assisted stack switching. Thus a fair bit
of this code is repurposing the MMU_SCRATCH reg for event handler
prologues to keep them re-entrant.

Many thanks to Rajeshwar Ranga for his initial "major" contributions to
SMP Port (back in 2008), and to Noam Camus and Gilad Ben-Yossef for help
with resurrecting that in 3.2 kernel (2012).

Note that this platform code is again singleton design pattern - so
multiple SMP platforms won't build at the moment - this deficiency is
addressed in subsequent patches within this series.

Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Rajeshwar Ranga <rajeshwar.ranga@gmail.com>
Cc: Noam Camus <noamc@ezchip.com>
Cc: Gilad Ben-Yossef <gilad@benyossef.com>

6 files changed, 181 insertions, 0 deletions

diff --git a/arch/arc/include/asm/entry.h b/arch/arc/include/asm/entry.h
index 23ef2de1e09..23daa326fc9 100644
--- a/arch/arc/include/asm/entry.h
+++ b/arch/arc/include/asm/entry.h
@@ -389,11 +389,19 @@
  * to be saved again on kernel mode stack, as part of ptregs.
  *-------------------------------------------------------------*/
 .macro EXCPN_PROLOG_FREEUP_REG	reg
+#ifdef CONFIG_SMP
+	sr  \reg, [ARC_REG_SCRATCH_DATA0]
+#else
 	st  \reg, [@ex_saved_reg1]
+#endif
 .endm
 
 .macro EXCPN_PROLOG_RESTORE_REG	reg
+#ifdef CONFIG_SMP
+	lr  \reg, [ARC_REG_SCRATCH_DATA0]
+#else
 	ld  \reg, [@ex_saved_reg1]
+#endif
 .endm
 
 /*--------------------------------------------------------------
@@ -508,7 +516,11 @@
 	/* restore original r9 , saved in int1_saved_reg
 	* It will be saved on stack in macro: SAVE_CALLER_SAVED
 	*/
+#ifdef CONFIG_SMP
+	lr  r9, [ARC_REG_SCRATCH_DATA0]
+#else
 	ld  r9, [@int1_saved_reg]
+#endif
 
 	/* now we are ready to save the remaining context :) */
 	st      orig_r8_IS_IRQ1, [sp, 8]    /* Event Type */
@@ -639,6 +651,41 @@
 	bmsk \reg, \reg, 7
 .endm
 
+#ifdef CONFIG_SMP
+
+/*-------------------------------------------------
+ * Retrieve the current running task on this CPU
+ * 1. Determine curr CPU id.
+ * 2. Use it to index into _current_task[ ]
+ */
+.macro  GET_CURR_TASK_ON_CPU   reg
+	GET_CPU_ID  \reg
+	ld.as  \reg, [@_current_task, \reg]
+.endm
+
+/*-------------------------------------------------
+ * Save a new task as the "current" task on this CPU
+ * 1. Determine curr CPU id.
+ * 2. Use it to index into _current_task[ ]
+ *
+ * Coded differently than GET_CURR_TASK_ON_CPU (which uses LD.AS)
+ * because ST r0, [r1, offset] can ONLY have s9 @offset
+ * while   LD can take s9 (4 byte insn) or LIMM (8 byte insn)
+ */
+
+.macro  SET_CURR_TASK_ON_CPU    tsk, tmp
+	GET_CPU_ID  \tmp
+	add2 \tmp, @_current_task, \tmp
+	st   \tsk, [\tmp]
+#ifdef CONFIG_ARC_CURR_IN_REG
+	mov r25, \tsk
+#endif
+
+.endm
+
+
+#else   /* Uniprocessor implementation of macros */
+
 .macro  GET_CURR_TASK_ON_CPU    reg
 	ld  \reg, [@_current_task]
 .endm
@@ -650,6 +697,8 @@
 #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
diff --git a/arch/arc/include/asm/mmu_context.h b/arch/arc/include/asm/mmu_context.h
index d12f3dec8b7..0d71fb11b57 100644
--- a/arch/arc/include/asm/mmu_context.h
+++ b/arch/arc/include/asm/mmu_context.h
@@ -147,8 +147,10 @@ init_new_context(struct task_struct *tsk, struct mm_struct *mm)
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 			     struct task_struct *tsk)
 {
+#ifndef CONFIG_SMP
 	/* PGD cached in MMU reg to avoid 3 mem lookups: task->mm->pgd */
 	write_aux_reg(ARC_REG_SCRATCH_DATA0, next->pgd);
+#endif
 
 	/*
 	 * Get a new ASID if task doesn't have a valid one. Possible when
@@ -197,7 +199,9 @@ static inline void destroy_context(struct mm_struct *mm)
 
 static inline void activate_mm(struct mm_struct *prev, struct mm_struct *next)
 {
+#ifndef CONFIG_SMP
 	write_aux_reg(ARC_REG_SCRATCH_DATA0, next->pgd);
+#endif
 
 	/* Unconditionally get a new ASID */
 	get_new_mmu_context(next);
diff --git a/arch/arc/include/asm/mutex.h b/arch/arc/include/asm/mutex.h
index 3be5e64da13..a2f88ff9f50 100644
--- a/arch/arc/include/asm/mutex.h
+++ b/arch/arc/include/asm/mutex.h
@@ -6,4 +6,13 @@
  * published by the Free Software Foundation.
  */
 
+/*
+ * xchg() based mutex fast path maintains a state of 0 or 1, as opposed to
+ * atomic dec based which can "count" any number of lock contenders.
+ * This ideally needs to be fixed in core, but for now switching to dec ver.
+ */
+#if defined(CONFIG_SMP) && (CONFIG_NR_CPUS > 2)
+#include <asm-generic/mutex-dec.h>
+#else
 #include <asm-generic/mutex-xchg.h>
+#endif
diff --git a/arch/arc/include/asm/pgtable.h b/arch/arc/include/asm/pgtable.h
index dcb0701528a..b7e36684c09 100644
--- a/arch/arc/include/asm/pgtable.h
+++ b/arch/arc/include/asm/pgtable.h
@@ -354,11 +354,15 @@ static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,
  * Thus use this macro only when you are certain that "current" is current
  * e.g. when dealing with signal frame setup code etc
  */
+#ifndef CONFIG_SMP
 #define pgd_offset_fast(mm, addr)	\
 ({					\
 	pgd_t *pgd_base = (pgd_t *) read_aux_reg(ARC_REG_SCRATCH_DATA0);  \
 	pgd_base + pgd_index(addr);	\
 })
+#else
+#define pgd_offset_fast(mm, addr)	pgd_offset(mm, addr)
+#endif
 
 extern void paging_init(void);
 extern pgd_t swapper_pg_dir[] __aligned(PAGE_SIZE);
diff --git a/arch/arc/include/asm/processor.h b/arch/arc/include/asm/processor.h
index b7b15561006..5f26b2c1cba 100644
--- a/arch/arc/include/asm/processor.h
+++ b/arch/arc/include/asm/processor.h
@@ -58,7 +58,15 @@ unsigned long thread_saved_pc(struct task_struct *t);
 /* Prepare to copy thread state - unlazy all lazy status */
 #define prepare_to_copy(tsk)    do { } while (0)
 
+/*
+ * A lot of busy-wait loops in SMP are based off of non-volatile data otherwise
+ * get optimised away by gcc
+ */
+#ifdef CONFIG_SMP
+#define cpu_relax()	__asm__ __volatile__ ("" : : : "memory")
+#else
 #define cpu_relax()	do { } while (0)
+#endif
 
 #define copy_segments(tsk, mm)      do { } while (0)
 #define release_segments(mm)        do { } while (0)
diff --git a/arch/arc/include/asm/smp.h b/arch/arc/include/asm/smp.h
index 4341f3ba7d9..f91f1946272 100644
--- a/arch/arc/include/asm/smp.h
+++ b/arch/arc/include/asm/smp.h
@@ -9,6 +9,69 @@
 #ifndef __ASM_ARC_SMP_H
 #define __ASM_ARC_SMP_H
 
+#ifdef CONFIG_SMP
+
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/threads.h>
+
+#define raw_smp_processor_id() (current_thread_info()->cpu)
+
+/* including cpumask.h leads to cyclic deps hence this Forward declaration */
+struct cpumask;
+
+/*
+ * APIs provided by arch SMP code to generic code
+ */
+extern void arch_send_call_function_single_ipi(int cpu);
+extern void arch_send_call_function_ipi_mask(const struct cpumask *mask);
+
+/*
+ * APIs provided by arch SMP code to rest of arch code
+ */
+extern void __init smp_init_cpus(void);
+extern void __init first_lines_of_secondary(void);
+
+/*
+ * API expected BY platform smp code (FROM arch smp code)
+ *
+ * smp_ipi_irq_setup:
+ *	Takes @cpu and @irq to which the arch-common ISR is hooked up
+ */
+extern int smp_ipi_irq_setup(int cpu, int irq);
+
+/*
+ * APIs expected FROM platform smp code
+ *
+ * arc_platform_smp_cpuinfo:
+ *	returns a string containing info for /proc/cpuinfo
+ *
+ * arc_platform_smp_init_cpu:
+ *	Called from start_kernel_secondary to do any CPU local setup
+ *	such as starting a timer, setting up IPI etc
+ *
+ * arc_platform_smp_wait_to_boot:
+ *	Called from early bootup code for non-Master CPUs to "park" them
+ *
+ * arc_platform_smp_wakeup_cpu:
+ *	Called from __cpu_up (Master CPU) to kick start another one
+ *
+ * arc_platform_ipi_send:
+ *	Takes @cpumask to which IPI(s) would be sent.
+ *	The actual msg-id/buffer is manager in arch-common code
+ *
+ * arc_platform_ipi_clear:
+ *	Takes @cpu which got IPI at @irq to do any IPI clearing
+ */
+extern const char *arc_platform_smp_cpuinfo(void);
+extern void arc_platform_smp_init_cpu(void);
+extern void arc_platform_smp_wait_to_boot(int cpu);
+extern void arc_platform_smp_wakeup_cpu(int cpu, unsigned long pc);
+extern void arc_platform_ipi_send(const struct cpumask *callmap);
+extern void arc_platform_ipi_clear(int cpu, int irq);
+
+#endif  /* CONFIG_SMP */
+
 /*
  * ARC700 doesn't support atomic Read-Modify-Write ops.
  * Originally Interrupts had to be disabled around code to gaurantee atomicity.
@@ -18,10 +81,52 @@
  *
  * (1) These insn were introduced only in 4.10 release. So for older released
  *	support needed.
+ *
+ * (2) In a SMP setup, the LLOCK/SCOND atomiticity across CPUs needs to be
+ *	gaurantted by the platform (not something which core handles).
+ *	Assuming a platform won't, SMP Linux needs to use spinlocks + local IRQ
+ *	disabling for atomicity.
+ *
+ *	However exported spinlock API is not usable due to cyclic hdr deps
+ *	(even after system.h disintegration upstream)
+ *	asm/bitops.h -> linux/spinlock.h -> linux/preempt.h
+ *		-> linux/thread_info.h -> linux/bitops.h -> asm/bitops.h
+ *
+ *	So the workaround is to use the lowest level arch spinlock API.
+ *	The exported spinlock API is smart enough to be NOP for !CONFIG_SMP,
+ *	but same is not true for ARCH backend, hence the need for 2 variants
  */
 #ifndef CONFIG_ARC_HAS_LLSC
 
 #include <linux/irqflags.h>
+#ifdef CONFIG_SMP
+
+#include <asm/spinlock.h>
+
+extern arch_spinlock_t smp_atomic_ops_lock;
+extern arch_spinlock_t smp_bitops_lock;
+
+#define atomic_ops_lock(flags)	do {		\
+	local_irq_save(flags);			\
+	arch_spin_lock(&smp_atomic_ops_lock);	\
+} while (0)
+
+#define atomic_ops_unlock(flags) do {		\
+	arch_spin_unlock(&smp_atomic_ops_lock);	\
+	local_irq_restore(flags);		\
+} while (0)
+
+#define bitops_lock(flags)	do {		\
+	local_irq_save(flags);			\
+	arch_spin_lock(&smp_bitops_lock);	\
+} while (0)
+
+#define bitops_unlock(flags) do {		\
+	arch_spin_unlock(&smp_bitops_lock);	\
+	local_irq_restore(flags);		\
+} while (0)
+
+#else /* !CONFIG_SMP */
 
 #define atomic_ops_lock(flags)		local_irq_save(flags)
 #define atomic_ops_unlock(flags)	local_irq_restore(flags)
@@ -29,6 +134,8 @@
 #define bitops_lock(flags)		local_irq_save(flags)
 #define bitops_unlock(flags)		local_irq_restore(flags)
 
+#endif /* !CONFIG_SMP */
+
 #endif	/* !CONFIG_ARC_HAS_LLSC */
 
 #endif