[PATCH 0/3] Use generic helper scu_power_mode()

[PATCH 0/3] Use generic helper scu_power_mode()

[Bastian Hecht]

Reposted to include the arch/arm mailing list.

The shmobile series implements its own code for setting the SCU power
register of the ARM MPCore. It uses 32-bit wide access and thus needs
locking as multiple CPUs might access it simultaneously for change. There is
already a small helper function that avoids the overhead by using 8-bit
wide access: As every CPU only accesses its own field we can drop the
lock and use it.

Bastian Hecht (3):
  ARM: shmobile: emev2: Replace modify_scu_cpu_psr with scu_power_mode
  ARM: shmobile: sh73a0: Replace modify_scu_cpu_psr with scu_power_mode
  ARM: shmobile: r8a7779: Replace modify_scu_cpu_psr with
    scu_power_mode

 arch/arm/mach-shmobile/smp-emev2.c   |   22 ++--------------------
 arch/arm/mach-shmobile/smp-r8a7779.c |   25 +++----------------------
 arch/arm/mach-shmobile/smp-sh73a0.c  |   23 ++---------------------
 3 files changed, 7 insertions(+), 63 deletions(-)

-- 
1.7.9.5

[PATCH 1/3] ARM: shmobile: emev2: Replace modify_scu_cpu_psr with scu_power_mode

[Bastian Hecht]

We can remove the extra code of modify_scu_cpu_psr() and use the cleaner
generic ARM helper scu_power_mode(). As every CPU only deals with its
own power register and scu_power_mode() operates with 8-bit accesses,
we save the locking overhead too.

Signed-off-by: Bastian Hecht <hechtb@gmail.com>
---
 arch/arm/mach-shmobile/smp-emev2.c |   22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)

diff --git a/arch/arm/mach-shmobile/smp-emev2.c b/arch/arm/mach-shmobile/smp-emev2.c
index f674562..535426c 100644
--- a/arch/arm/mach-shmobile/smp-emev2.c
+++ b/arch/arm/mach-shmobile/smp-emev2.c
@@ -32,24 +32,8 @@
 
 #define EMEV2_SCU_BASE 0x1e000000
 
-static DEFINE_SPINLOCK(scu_lock);
 static void __iomem *scu_base;
 
-static void modify_scu_cpu_psr(unsigned long set, unsigned long clr)
-{
-	unsigned long tmp;
-
-	/* we assume this code is running on a different cpu
-	 * than the one that is changing coherency setting */
-	spin_lock(&scu_lock);
-	tmp = readl(scu_base + 8);
-	tmp &= ~clr;
-	tmp |= set;
-	writel(tmp, scu_base + 8);
-	spin_unlock(&scu_lock);
-
-}
-
 static unsigned int __init emev2_get_core_count(void)
 {
 	if (!scu_base) {
@@ -95,7 +79,7 @@ static int __cpuinit emev2_boot_secondary(unsigned int cpu, struct task_struct *
 	cpu = cpu_logical_map(cpu);
 
 	/* enable cache coherency */
-	modify_scu_cpu_psr(0, 3 << (cpu * 8));
+	scu_power_mode(scu_base, 0);
 
 	/* Tell ROM loader about our vector (in headsmp.S) */
 	emev2_set_boot_vector(__pa(shmobile_secondary_vector));
@@ -106,12 +90,10 @@ static int __cpuinit emev2_boot_secondary(unsigned int cpu, struct task_struct *
 
 static void __init emev2_smp_prepare_cpus(unsigned int max_cpus)
 {
-	int cpu = cpu_logical_map(0);
-
 	scu_enable(scu_base);
 
 	/* enable cache coherency on CPU0 */
-	modify_scu_cpu_psr(0, 3 << (cpu * 8));
+	scu_power_mode(scu_base, 0);
 }
 
 static void __init emev2_smp_init_cpus(void)
-- 
1.7.9.5

[PATCH 2/3] ARM: shmobile: sh73a0: Replace modify_scu_cpu_psr with scu_power_mode

[Bastian Hecht]

We can remove the extra code of modify_scu_cpu_psr() and use the cleaner
generic ARM helper scu_power_mode(). As every CPU only deals with its
own power register and scu_power_mode() operates with 8-bit accesses,
we save the locking overhead too.

Signed-off-by: Bastian Hecht <hechtb@gmail.com>
---
 arch/arm/mach-shmobile/smp-sh73a0.c |   23 ++---------------------
 1 file changed, 2 insertions(+), 21 deletions(-)

diff --git a/arch/arm/mach-shmobile/smp-sh73a0.c b/arch/arm/mach-shmobile/smp-sh73a0.c
index 624f00f..96ddb97 100644
--- a/arch/arm/mach-shmobile/smp-sh73a0.c
+++ b/arch/arm/mach-shmobile/smp-sh73a0.c
@@ -41,9 +41,6 @@ static void __iomem *scu_base_addr(void)
 	return (void __iomem *)0xf0000000;
 }
 
-static DEFINE_SPINLOCK(scu_lock);
-static unsigned long tmp;
-
 #ifdef CONFIG_HAVE_ARM_TWD
 static DEFINE_TWD_LOCAL_TIMER(twd_local_timer, 0xf0000600, 29);
 void __init sh73a0_register_twd(void)
@@ -52,20 +49,6 @@ void __init sh73a0_register_twd(void)
 }
 #endif
 
-static void modify_scu_cpu_psr(unsigned long set, unsigned long clr)
-{
-	void __iomem *scu_base = scu_base_addr();
-
-	spin_lock(&scu_lock);
-	tmp = __raw_readl(scu_base + 8);
-	tmp &= ~clr;
-	tmp |= set;
-	spin_unlock(&scu_lock);
-
-	/* disable cache coherency after releasing the lock */
-	__raw_writel(tmp, scu_base + 8);
-}
-
 static unsigned int __init sh73a0_get_core_count(void)
 {
 	void __iomem *scu_base = scu_base_addr();
@@ -83,7 +66,7 @@ static int __cpuinit sh73a0_boot_secondary(unsigned int cpu, struct task_struct
 	cpu = cpu_logical_map(cpu);
 
 	/* enable cache coherency */
-	modify_scu_cpu_psr(0, 3 << (cpu * 8));
+	scu_power_mode(scu_base_addr(), 0);
 
 	if (((__raw_readl(PSTR) >> (4 * cpu)) & 3) == 3)
 		__raw_writel(1 << cpu, WUPCR);	/* wake up */
@@ -95,8 +78,6 @@ static int __cpuinit sh73a0_boot_secondary(unsigned int cpu, struct task_struct
 
 static void __init sh73a0_smp_prepare_cpus(unsigned int max_cpus)
 {
-	int cpu = cpu_logical_map(0);
-
 	scu_enable(scu_base_addr());
 
 	/* Map the reset vector (in headsmp.S) */
@@ -104,7 +85,7 @@ static void __init sh73a0_smp_prepare_cpus(unsigned int max_cpus)
 	__raw_writel(__pa(shmobile_secondary_vector), SBAR);
 
 	/* enable cache coherency on CPU0 */
-	modify_scu_cpu_psr(0, 3 << (cpu * 8));
+	scu_power_mode(scu_base_addr(), 0);
 }
 
 static void __init sh73a0_smp_init_cpus(void)
-- 
1.7.9.5

[PATCH 3/3] ARM: shmobile: r8a7779: Replace modify_scu_cpu_psr with scu_power_mode

[Bastian Hecht]

We can remove the extra code of modify_scu_cpu_psr() and use the cleaner
generic ARM helper scu_power_mode(). As every CPU only deals with its
own power register and scu_power_mode() operates with 8-bit accesses,
we save the locking overhead too.

Signed-off-by: Bastian Hecht <hechtb@gmail.com>
---
 arch/arm/mach-shmobile/smp-r8a7779.c |   25 +++----------------------
 1 file changed, 3 insertions(+), 22 deletions(-)

diff --git a/arch/arm/mach-shmobile/smp-r8a7779.c b/arch/arm/mach-shmobile/smp-r8a7779.c
index 2ce6af9..9def0f2 100644
--- a/arch/arm/mach-shmobile/smp-r8a7779.c
+++ b/arch/arm/mach-shmobile/smp-r8a7779.c
@@ -61,9 +61,6 @@ static void __iomem *scu_base_addr(void)
 	return (void __iomem *)0xf0000000;
 }
 
-static DEFINE_SPINLOCK(scu_lock);
-static unsigned long tmp;
-
 #ifdef CONFIG_HAVE_ARM_TWD
 static DEFINE_TWD_LOCAL_TIMER(twd_local_timer, 0xf0000600, 29);
 
@@ -73,20 +70,6 @@ void __init r8a7779_register_twd(void)
 }
 #endif
 
-static void modify_scu_cpu_psr(unsigned long set, unsigned long clr)
-{
-	void __iomem *scu_base = scu_base_addr();
-
-	spin_lock(&scu_lock);
-	tmp = __raw_readl(scu_base + 8);
-	tmp &= ~clr;
-	tmp |= set;
-	spin_unlock(&scu_lock);
-
-	/* disable cache coherency after releasing the lock */
-	__raw_writel(tmp, scu_base + 8);
-}
-
 static unsigned int __init r8a7779_get_core_count(void)
 {
 	void __iomem *scu_base = scu_base_addr();
@@ -102,7 +85,7 @@ static int r8a7779_platform_cpu_kill(unsigned int cpu)
 	cpu = cpu_logical_map(cpu);
 
 	/* disable cache coherency */
-	modify_scu_cpu_psr(3 << (cpu * 8), 0);
+	scu_power_mode(scu_base_addr(), 3);
 
 	if (cpu < ARRAY_SIZE(r8a7779_ch_cpu))
 		ch = r8a7779_ch_cpu[cpu];
@@ -145,7 +128,7 @@ static int __cpuinit r8a7779_boot_secondary(unsigned int cpu, struct task_struct
 	cpu = cpu_logical_map(cpu);
 
 	/* enable cache coherency */
-	modify_scu_cpu_psr(0, 3 << (cpu * 8));
+	scu_power_mode(scu_base_addr(), 0);
 
 	if (cpu < ARRAY_SIZE(r8a7779_ch_cpu))
 		ch = r8a7779_ch_cpu[cpu];
@@ -158,15 +141,13 @@ static int __cpuinit r8a7779_boot_secondary(unsigned int cpu, struct task_struct
 
 static void __init r8a7779_smp_prepare_cpus(unsigned int max_cpus)
 {
-	int cpu = cpu_logical_map(0);
-
 	scu_enable(scu_base_addr());
 
 	/* Map the reset vector (in headsmp.S) */
 	__raw_writel(__pa(shmobile_secondary_vector), AVECR);
 
 	/* enable cache coherency on CPU0 */
-	modify_scu_cpu_psr(0, 3 << (cpu * 8));
+	scu_power_mode(scu_base_addr(), 0);
 
 	r8a7779_pm_init();
 
-- 
1.7.9.5

[PATCH 0/3] Use generic helper scu_power_mode()

[Simon Horman]

On Thu, Oct 25, 2012 at 12:58:19PM +0200, Bastian Hecht wrote:
> Reposted to include the arch/arm mailing list.
> 
> The shmobile series implements its own code for setting the SCU power
> register of the ARM MPCore. It uses 32-bit wide access and thus needs
> locking as multiple CPUs might access it simultaneously for change. There is
> already a small helper function that avoids the overhead by using 8-bit
> wide access: As every CPU only accesses its own field we can drop the
> lock and use it.

Thanks,

these look good to me. I will push them to the boards branch once
its current pull request has been processed by the arm-soc people.