From 9a6655e49fd98f3748bb80da20705448aad9ee57 Mon Sep 17 00:00:00 2001 From: Catalin Marinas Date: Tue, 31 Aug 2010 13:05:22 +0100 Subject: ARM: Improve the L2 cache performance when PL310 is used With this L2 cache controller, the cache maintenance by PA and sync operations are atomic and do not require a "wait" loop. This patch conditionally defines the cache_wait() function. Since L2x0 cache controllers do not work with ARMv7 CPUs, the patch automatically enables CACHE_PL310 when only CPU_V7 is defined. Signed-off-by: Catalin Marinas --- arch/arm/mm/Kconfig | 8 ++++++++ arch/arm/mm/cache-l2x0.c | 15 ++++++++++++--- 2 files changed, 20 insertions(+), 3 deletions(-) (limited to 'arch/arm/mm') diff --git a/arch/arm/mm/Kconfig b/arch/arm/mm/Kconfig index a0a2928..4414a01 100644 --- a/arch/arm/mm/Kconfig +++ b/arch/arm/mm/Kconfig @@ -779,6 +779,14 @@ config CACHE_L2X0 help This option enables the L2x0 PrimeCell. +config CACHE_PL310 + bool + depends on CACHE_L2X0 + default y if CPU_V7 && !CPU_V6 + help + This option enables optimisations for the PL310 cache + controller. + config CACHE_TAUROS2 bool "Enable the Tauros2 L2 cache controller" depends on (ARCH_DOVE || ARCH_MMP) diff --git a/arch/arm/mm/cache-l2x0.c b/arch/arm/mm/cache-l2x0.c index 9982eb3..edb43ff 100644 --- a/arch/arm/mm/cache-l2x0.c +++ b/arch/arm/mm/cache-l2x0.c @@ -29,13 +29,22 @@ static void __iomem *l2x0_base; static DEFINE_SPINLOCK(l2x0_lock); static uint32_t l2x0_way_mask; /* Bitmask of active ways */ -static inline void cache_wait(void __iomem *reg, unsigned long mask) +static inline void cache_wait_way(void __iomem *reg, unsigned long mask) { - /* wait for the operation to complete */ + /* wait for cache operation by line or way to complete */ while (readl_relaxed(reg) & mask) ; } +#ifdef CONFIG_CACHE_PL310 +static inline void cache_wait(void __iomem *reg, unsigned long mask) +{ + /* cache operations by line are atomic on PL310 */ +} +#else +#define cache_wait cache_wait_way +#endif + static inline void cache_sync(void) { void __iomem *base = l2x0_base; @@ -110,7 +119,7 @@ static inline void l2x0_inv_all(void) /* invalidate all ways */ spin_lock_irqsave(&l2x0_lock, flags); writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_INV_WAY); - cache_wait(l2x0_base + L2X0_INV_WAY, l2x0_way_mask); + cache_wait_way(l2x0_base + L2X0_INV_WAY, l2x0_way_mask); cache_sync(); spin_unlock_irqrestore(&l2x0_lock, flags); } -- cgit v1.1 From 2fd8658931193599c867fd6974fa184ec34af16c Mon Sep 17 00:00:00 2001 From: Thomas Gleixner <[tglx@linutronix.de]> Date: Sat, 31 Jul 2010 21:05:24 +0530 Subject: arm: Implement l2x0 cache disable functions Add flush_all, inv_all and disable functions to the l2x0 code. These functions are called from kexec code to prevent random crashes in the new kernel. Platforms like OMAP which control L2 enable/disable via SMI mode can override the outer_cache.disable() function to implement their own. Signed-off-by: Thomas Gleixner Acked-by: Catalin Marinas Acked-by: Linus Walleij --- arch/arm/mm/cache-l2x0.c | 28 +++++++++++++++++++++++++++- 1 file changed, 27 insertions(+), 1 deletion(-) (limited to 'arch/arm/mm') diff --git a/arch/arm/mm/cache-l2x0.c b/arch/arm/mm/cache-l2x0.c index edb43ff..9310d61 100644 --- a/arch/arm/mm/cache-l2x0.c +++ b/arch/arm/mm/cache-l2x0.c @@ -112,12 +112,26 @@ static void l2x0_cache_sync(void) spin_unlock_irqrestore(&l2x0_lock, flags); } -static inline void l2x0_inv_all(void) +static void l2x0_flush_all(void) +{ + unsigned long flags; + + /* clean all ways */ + spin_lock_irqsave(&l2x0_lock, flags); + writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_CLEAN_INV_WAY); + cache_wait_way(l2x0_base + L2X0_CLEAN_INV_WAY, l2x0_way_mask); + cache_sync(); + spin_unlock_irqrestore(&l2x0_lock, flags); +} + +static void l2x0_inv_all(void) { unsigned long flags; /* invalidate all ways */ spin_lock_irqsave(&l2x0_lock, flags); + /* Invalidating when L2 is enabled is a nono */ + BUG_ON(readl(l2x0_base + L2X0_CTRL) & 1); writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_INV_WAY); cache_wait_way(l2x0_base + L2X0_INV_WAY, l2x0_way_mask); cache_sync(); @@ -215,6 +229,15 @@ static void l2x0_flush_range(unsigned long start, unsigned long end) spin_unlock_irqrestore(&l2x0_lock, flags); } +static void l2x0_disable(void) +{ + unsigned long flags; + + spin_lock_irqsave(&l2x0_lock, flags); + writel(0, l2x0_base + L2X0_CTRL); + spin_unlock_irqrestore(&l2x0_lock, flags); +} + void __init l2x0_init(void __iomem *base, __u32 aux_val, __u32 aux_mask) { __u32 aux; @@ -272,6 +295,9 @@ void __init l2x0_init(void __iomem *base, __u32 aux_val, __u32 aux_mask) outer_cache.clean_range = l2x0_clean_range; outer_cache.flush_range = l2x0_flush_range; outer_cache.sync = l2x0_cache_sync; + outer_cache.flush_all = l2x0_flush_all; + outer_cache.inv_all = l2x0_inv_all; + outer_cache.disable = l2x0_disable; printk(KERN_INFO "%s cache controller enabled\n", type); printk(KERN_INFO "l2x0: %d ways, CACHE_ID 0x%08x, AUX_CTRL 0x%08x\n", -- cgit v1.1 From 5ba70372289a1fb378b95cee2cf46b0203d65291 Mon Sep 17 00:00:00 2001 From: Santosh Shilimkar Date: Sun, 11 Jul 2010 14:35:37 +0530 Subject: ARM: l2x0: Determine the cache size The cache size is needed for to optimise range based maintainance operations Signed-off-by: Santosh Shilimkar Acked-by: Catalin Marinas Acked-by: Linus Walleij --- arch/arm/mm/cache-l2x0.c | 13 +++++++++++-- 1 file changed, 11 insertions(+), 2 deletions(-) (limited to 'arch/arm/mm') diff --git a/arch/arm/mm/cache-l2x0.c b/arch/arm/mm/cache-l2x0.c index 9310d61..262c752 100644 --- a/arch/arm/mm/cache-l2x0.c +++ b/arch/arm/mm/cache-l2x0.c @@ -28,6 +28,7 @@ static void __iomem *l2x0_base; static DEFINE_SPINLOCK(l2x0_lock); static uint32_t l2x0_way_mask; /* Bitmask of active ways */ +static uint32_t l2x0_size; static inline void cache_wait_way(void __iomem *reg, unsigned long mask) { @@ -242,6 +243,7 @@ void __init l2x0_init(void __iomem *base, __u32 aux_val, __u32 aux_mask) { __u32 aux; __u32 cache_id; + __u32 way_size = 0; int ways; const char *type; @@ -276,6 +278,13 @@ void __init l2x0_init(void __iomem *base, __u32 aux_val, __u32 aux_mask) l2x0_way_mask = (1 << ways) - 1; /* + * L2 cache Size = Way size * Number of ways + */ + way_size = (aux & L2X0_AUX_CTRL_WAY_SIZE_MASK) >> 17; + way_size = 1 << (way_size + 3); + l2x0_size = ways * way_size * SZ_1K; + + /* * Check if l2x0 controller is already enabled. * If you are booting from non-secure mode * accessing the below registers will fault. @@ -300,6 +309,6 @@ void __init l2x0_init(void __iomem *base, __u32 aux_val, __u32 aux_mask) outer_cache.disable = l2x0_disable; printk(KERN_INFO "%s cache controller enabled\n", type); - printk(KERN_INFO "l2x0: %d ways, CACHE_ID 0x%08x, AUX_CTRL 0x%08x\n", - ways, cache_id, aux); + printk(KERN_INFO "l2x0: %d ways, CACHE_ID 0x%08x, AUX_CTRL 0x%08x, Cache size: %d B\n", + ways, cache_id, aux, l2x0_size); } -- cgit v1.1 From 444457c1f59d58bc48acf5b4fc585225106c11ff Mon Sep 17 00:00:00 2001 From: Santosh Shilimkar Date: Sun, 11 Jul 2010 14:58:41 +0530 Subject: ARM: l2x0: Optimise the range based operations For the big buffers which are in excess of cache size, the maintaince operations by PA are very slow. For such buffers the maintainace operations can be speeded up by using the WAY based method. Signed-off-by: Santosh Shilimkar Acked-by: Catalin Marinas Acked-by: Linus Walleij --- arch/arm/mm/cache-l2x0.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'arch/arm/mm') diff --git a/arch/arm/mm/cache-l2x0.c b/arch/arm/mm/cache-l2x0.c index 262c752..170c9bb 100644 --- a/arch/arm/mm/cache-l2x0.c +++ b/arch/arm/mm/cache-l2x0.c @@ -125,6 +125,18 @@ static void l2x0_flush_all(void) spin_unlock_irqrestore(&l2x0_lock, flags); } +static void l2x0_clean_all(void) +{ + unsigned long flags; + + /* clean all ways */ + spin_lock_irqsave(&l2x0_lock, flags); + writel_relaxed(l2x0_way_mask, l2x0_base + L2X0_CLEAN_WAY); + cache_wait_way(l2x0_base + L2X0_CLEAN_WAY, l2x0_way_mask); + cache_sync(); + spin_unlock_irqrestore(&l2x0_lock, flags); +} + static void l2x0_inv_all(void) { unsigned long flags; @@ -183,6 +195,11 @@ static void l2x0_clean_range(unsigned long start, unsigned long end) void __iomem *base = l2x0_base; unsigned long flags; + if ((end - start) >= l2x0_size) { + l2x0_clean_all(); + return; + } + spin_lock_irqsave(&l2x0_lock, flags); start &= ~(CACHE_LINE_SIZE - 1); while (start < end) { @@ -208,6 +225,11 @@ static void l2x0_flush_range(unsigned long start, unsigned long end) void __iomem *base = l2x0_base; unsigned long flags; + if ((end - start) >= l2x0_size) { + l2x0_flush_all(); + return; + } + spin_lock_irqsave(&l2x0_lock, flags); start &= ~(CACHE_LINE_SIZE - 1); while (start < end) { -- cgit v1.1