diff mbox

[8/9] ARM: kernel: add support for cpu cache information

Message ID 1403717444-23559-9-git-send-email-sudeep.holla@arm.com
State New
Headers show

Commit Message

Sudeep Holla June 25, 2014, 5:30 p.m. UTC
From: Sudeep Holla <sudeep.holla@arm.com>

This patch adds support for cacheinfo on ARM platforms.

On ARMv7, the cache hierarchy can be identified through Cache Level ID
register(CLIDR) while the cache geometry is provided by Cache Size ID
register(CCSIDR).

On architecture versions before ARMv7, CLIDR and CCSIDR is not
implemented. The cache type register(CTR) provides both cache hierarchy
and geometry if implemented. For implementations that doesn't support
CTR, we need to list the probable value of CTR if it was implemented
along with the cpuid for the sake of simplicity to handle them.

Since the architecture doesn't provide any way of detecting the cpus
sharing particular cache, device tree is used fo the same purpose.
On non-DT platforms, first level caches are per-cpu while higher level
caches are assumed system-wide.

Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Nicolas Pitre <nicolas.pitre@linaro.org>
Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: linux-arm-kernel@lists.infradead.org
---
 arch/arm/kernel/Makefile    |   1 +
 arch/arm/kernel/cacheinfo.c | 229 ++++++++++++++++++++++++++++++++++++++++++++
 arch/arm/mm/Kconfig         |  13 +++
 3 files changed, 243 insertions(+)
 create mode 100644 arch/arm/kernel/cacheinfo.c

Comments

Russell King - ARM Linux June 25, 2014, 10:33 p.m. UTC | #1
On Wed, Jun 25, 2014 at 06:30:43PM +0100, Sudeep Holla wrote:
> diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
> index 38ddd9f..2c5ff0e 100644
> --- a/arch/arm/kernel/Makefile
> +++ b/arch/arm/kernel/Makefile
> @@ -29,6 +29,7 @@ obj-y		+= entry-v7m.o v7m.o
>  else
>  obj-y		+= entry-armv.o
>  endif
> +obj-$(CONFIG_CPU_HAS_CACHE) += cacheinfo.o
>  
>  obj-$(CONFIG_OC_ETM)		+= etm.o
>  obj-$(CONFIG_CPU_IDLE)		+= cpuidle.o
> diff --git a/arch/arm/kernel/cacheinfo.c b/arch/arm/kernel/cacheinfo.c
> new file mode 100644
> index 0000000..ab70993
> --- /dev/null
> +++ b/arch/arm/kernel/cacheinfo.c
> @@ -0,0 +1,229 @@
> +/*
> + *  ARM cacheinfo support
> + *
> + *  Copyright (C) 2014 ARM Ltd.
> + *  All Rights Reserved
> + *
> + * 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/bitops.h>
> +#include <linux/cacheinfo.h>
> +#include <linux/cpu.h>
> +#include <linux/compiler.h>
> +#include <linux/of.h>
> +
> +#include <asm/cputype.h>
> +#include <asm/processor.h>
> +
> +#if __LINUX_ARM_ARCH__ < 7 /* pre ARMv7 */

__LINUX_ARM_ARCH__ defines the minimum architecture version we are building
for - we may support later versions than the architecture version denoted
by this symbol.  It does not define which CPUs we are building for.  Are
you sure that this is correct here?  What if we build a kernel supporting
both v6 + v7, as the OMAP guys do?

> +
> +#define MAX_CACHE_LEVEL		1	/* Only 1 level supported */
> +#define CTR_CTYPE_SHIFT		24
> +#define CTR_CTYPE_MASK		(1 << CTR_CTYPE_SHIFT)
> +
> +struct ctr_info {
> +	unsigned int cpuid_id;
> +	unsigned int ctr;
> +};
> +
> +static struct ctr_info cache_ctr_list[] = {
> +};

This list needs to be populated.  Early CPUs (such as StrongARM) do not
have the CTR register.

> +static int get_unimplemented_ctr(unsigned int *ctr)
> +{
> +	int i, cpuid_id = read_cpuid_id();
> +
> +	for (i = 0; i < ARRAY_SIZE(cache_ctr_list); i++)
> +		if (cache_ctr_list[i].cpuid_id == cpuid_id) {
> +			*ctr = cache_ctr_list[i].ctr;
> +			return 0;
> +		}
> +	return -ENOENT;
> +}
> +
> +static unsigned int get_ctr(void)
> +{
> +	unsigned int ctr;
> +
> +	if (get_unimplemented_ctr(&ctr))
> +		asm volatile ("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));

read_cpuid_cachetype() ?
Stephen Boyd June 26, 2014, 12:19 a.m. UTC | #2
On 06/25/14 10:30, Sudeep Holla wrote:
> +
> +/*
> + * Which cache CCSIDR represents depends on CSSELR value
> + * Make sure no one else changes CSSELR during this
> + * smp_call_function_single prevents preemption for us
> + */

Where's the smp_call_function_single() or preemption disable happening?

> +static inline u32 get_ccsidr(u32 csselr)
> +{
> +	u32 ccsidr;
> +
> +	/* Put value into CSSELR */
> +	asm volatile ("mcr p15, 2, %0, c0, c0, 0" : : "r" (csselr));
> +	isb();
> +	/* Read result out of CCSIDR */
> +	asm volatile ("mrc p15, 1, %0, c0, c0, 0" : "=r" (ccsidr));
> +
> +	return ccsidr;
> +}
> +
Sudeep Holla June 26, 2014, 11:33 a.m. UTC | #3
Hi Russell,

Thanks for the reviews.

On 25/06/14 23:33, Russell King - ARM Linux wrote:
> On Wed, Jun 25, 2014 at 06:30:43PM +0100, Sudeep Holla wrote:
[...]
>> +
>> +#include <linux/bitops.h>
>> +#include <linux/cacheinfo.h>
>> +#include <linux/cpu.h>
>> +#include <linux/compiler.h>
>> +#include <linux/of.h>
>> +
>> +#include <asm/cputype.h>
>> +#include <asm/processor.h>
>> +
>> +#if __LINUX_ARM_ARCH__ < 7 /* pre ARMv7 */
>
> __LINUX_ARM_ARCH__ defines the minimum architecture version we are building
> for - we may support later versions than the architecture version denoted
> by this symbol.  It does not define which CPUs we are building for.  Are
> you sure that this is correct here?  What if we build a kernel supporting
> both v6 + v7, as the OMAP guys do?
>

You are right, I have not considered v6 + v7, I will use cpu_architecture and
make it runtime.

>> +
>> +#define MAX_CACHE_LEVEL		1	/* Only 1 level supported */
>> +#define CTR_CTYPE_SHIFT		24
>> +#define CTR_CTYPE_MASK		(1 << CTR_CTYPE_SHIFT)
>> +
>> +struct ctr_info {
>> +	unsigned int cpuid_id;
>> +	unsigned int ctr;
>> +};
>> +
>> +static struct ctr_info cache_ctr_list[] = {
>> +};
>
> This list needs to be populated.  Early CPUs (such as StrongARM) do not
> have the CTR register.
>

Right, since I didn't have the list left it empty. I will compile the list
soon but I need your help. The list of StrongARM I can come up is:
1. SA-110
2. SA-1100
3. SA-1110
4. SA-1500 (grep didn't show this in kernel, not sure if it's supported)

I also have to find all other ARMv4 implementations not having CTR.

>> +static int get_unimplemented_ctr(unsigned int *ctr)
>> +{
>> +	int i, cpuid_id = read_cpuid_id();
>> +
>> +	for (i = 0; i < ARRAY_SIZE(cache_ctr_list); i++)
>> +		if (cache_ctr_list[i].cpuid_id == cpuid_id) {
>> +			*ctr = cache_ctr_list[i].ctr;
>> +			return 0;
>> +		}
>> +	return -ENOENT;
>> +}
>> +
>> +static unsigned int get_ctr(void)
>> +{
>> +	unsigned int ctr;
>> +
>> +	if (get_unimplemented_ctr(&ctr))
>> +		asm volatile ("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
>
> read_cpuid_cachetype() ?
>
Ah, I missed to see that, will use it.

Regards,
Sudeep

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Sudeep Holla June 26, 2014, 11:36 a.m. UTC | #4
Hi Stephen,

On 26/06/14 01:19, Stephen Boyd wrote:
> On 06/25/14 10:30, Sudeep Holla wrote:
>> +
>> +/*
>> + * Which cache CCSIDR represents depends on CSSELR value
>> + * Make sure no one else changes CSSELR during this
>> + * smp_call_function_single prevents preemption for us
>> + */
>
> Where's the smp_call_function_single() or preemption disable happening?
>

init_cache_level is called using smp_call_function_single in
drivers/base/cacheinfo.c(PATCH 2/9)

Regards,
Sudeep

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
Stephen Boyd June 26, 2014, 6:45 p.m. UTC | #5
On 06/26/14 04:36, Sudeep Holla wrote:
> Hi Stephen,
>
> On 26/06/14 01:19, Stephen Boyd wrote:
>> On 06/25/14 10:30, Sudeep Holla wrote:
>>> +
>>> +/*
>>> + * Which cache CCSIDR represents depends on CSSELR value
>>> + * Make sure no one else changes CSSELR during this
>>> + * smp_call_function_single prevents preemption for us
>>> + */
>>
>> Where's the smp_call_function_single() or preemption disable happening?
>>
>
> init_cache_level is called using smp_call_function_single in
> drivers/base/cacheinfo.c(PATCH 2/9)

Oh that's unexpected. Do other architectures require the use of
smp_call_function_single() to read their cache information? It seems
like an ARM architecture specific detail that has been pushed up into
the generic layer.
Sudeep Holla June 27, 2014, 9:38 a.m. UTC | #6
Hi,

On 26/06/14 19:45, Stephen Boyd wrote:
> On 06/26/14 04:36, Sudeep Holla wrote:
>> Hi Stephen,
>>
>> On 26/06/14 01:19, Stephen Boyd wrote:
>>> On 06/25/14 10:30, Sudeep Holla wrote:
>>>> +
>>>> +/*
>>>> + * Which cache CCSIDR represents depends on CSSELR value
>>>> + * Make sure no one else changes CSSELR during this
>>>> + * smp_call_function_single prevents preemption for us
>>>> + */
>>>
>>> Where's the smp_call_function_single() or preemption disable happening?
>>>
>>
>> init_cache_level is called using smp_call_function_single in
>> drivers/base/cacheinfo.c(PATCH 2/9)
>
> Oh that's unexpected. Do other architectures require the use of
> smp_call_function_single() to read their cache information? It seems
> like an ARM architecture specific detail that has been pushed up into
> the generic layer.
>

Right, since I started with x86 as reference and it requires it, I missed to
see others. So x86,ARM{32,64} requires it while ppc,s390 and ia64 doesn't.
I see how to fix that. Thanks for spotting it.

Regards,
Sudeep

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
diff mbox

Patch

diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 38ddd9f..2c5ff0e 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -29,6 +29,7 @@  obj-y		+= entry-v7m.o v7m.o
 else
 obj-y		+= entry-armv.o
 endif
+obj-$(CONFIG_CPU_HAS_CACHE) += cacheinfo.o
 
 obj-$(CONFIG_OC_ETM)		+= etm.o
 obj-$(CONFIG_CPU_IDLE)		+= cpuidle.o
diff --git a/arch/arm/kernel/cacheinfo.c b/arch/arm/kernel/cacheinfo.c
new file mode 100644
index 0000000..ab70993
--- /dev/null
+++ b/arch/arm/kernel/cacheinfo.c
@@ -0,0 +1,229 @@ 
+/*
+ *  ARM cacheinfo support
+ *
+ *  Copyright (C) 2014 ARM Ltd.
+ *  All Rights Reserved
+ *
+ * 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/bitops.h>
+#include <linux/cacheinfo.h>
+#include <linux/cpu.h>
+#include <linux/compiler.h>
+#include <linux/of.h>
+
+#include <asm/cputype.h>
+#include <asm/processor.h>
+
+#if __LINUX_ARM_ARCH__ < 7 /* pre ARMv7 */
+
+#define MAX_CACHE_LEVEL		1	/* Only 1 level supported */
+#define CTR_CTYPE_SHIFT		24
+#define CTR_CTYPE_MASK		(1 << CTR_CTYPE_SHIFT)
+
+struct ctr_info {
+	unsigned int cpuid_id;
+	unsigned int ctr;
+};
+
+static struct ctr_info cache_ctr_list[] = {
+};
+
+static int get_unimplemented_ctr(unsigned int *ctr)
+{
+	int i, cpuid_id = read_cpuid_id();
+
+	for (i = 0; i < ARRAY_SIZE(cache_ctr_list); i++)
+		if (cache_ctr_list[i].cpuid_id == cpuid_id) {
+			*ctr = cache_ctr_list[i].ctr;
+			return 0;
+		}
+	return -ENOENT;
+}
+
+static unsigned int get_ctr(void)
+{
+	unsigned int ctr;
+
+	if (get_unimplemented_ctr(&ctr))
+		asm volatile ("mrc p15, 0, %0, c0, c0, 1" : "=r" (ctr));
+	return ctr;
+}
+
+static enum cache_type get_cache_type(int level)
+{
+	if (level > MAX_CACHE_LEVEL)
+		return CACHE_TYPE_NOCACHE;
+	return get_ctr() & CTR_CTYPE_MASK ?
+		CACHE_TYPE_SEPARATE : CACHE_TYPE_UNIFIED;
+}
+
+/*
+ *  +---------------------------------+
+ *  | 9  8  7  6 | 5  4  3 | 2 | 1  0 |
+ *  +---------------------------------+
+ *  |    size    |  assoc  | m |  len |
+ *  +---------------------------------+
+ * linelen        = 1 << (len + 3)
+ * multiplier     = 2 + m
+ * nsets          = 1 << (size + 6 - assoc - len)
+ * associativity  = multiplier << (assoc - 1)
+ * cache_size     = multiplier << (size + 8)
+ */
+#define CTR_LINESIZE_MASK	0x3
+#define CTR_MULTIPLIER_SHIFT	2
+#define CTR_MULTIPLIER_MASK	0x1
+#define CTR_ASSOCIAT_SHIFT	3
+#define CTR_ASSOCIAT_MASK	0x7
+#define CTR_SIZE_SHIFT		6
+#define CTR_SIZE_MASK		0xF
+#define CTR_DCACHE_SHIFT	12
+
+static void __ci_leaf_init(enum cache_type type, struct cacheinfo *this_leaf)
+{
+	unsigned int size, multiplier, assoc, len, tmp = get_ctr();
+
+	if (type == CACHE_TYPE_DATA)
+		tmp >>= CTR_DCACHE_SHIFT;
+
+	len = tmp & CTR_LINESIZE_MASK;
+	size = (tmp >> CTR_SIZE_SHIFT) & CTR_SIZE_MASK;
+	assoc = (tmp >> CTR_ASSOCIAT_SHIFT) & CTR_ASSOCIAT_MASK;
+	multiplier = ((tmp >> CTR_MULTIPLIER_SHIFT) & CTR_MULTIPLIER_MASK) + 2;
+
+	this_leaf->type = type;
+	this_leaf->coherency_line_size = 1 << (len + 3);
+	this_leaf->number_of_sets = 1 << (size + 6 - assoc - len);
+	this_leaf->ways_of_associativity = multiplier << (assoc - 1);
+	this_leaf->size = multiplier << (size + 8);
+}
+
+#else /* ARMv7 */
+
+#define MAX_CACHE_LEVEL			7	/* Max 7 level supported */
+/* Ctypen, bits[3(n - 1) + 2 : 3(n - 1)], for n = 1 to 7 */
+#define CLIDR_CTYPE_SHIFT(level)	(3 * (level - 1))
+#define CLIDR_CTYPE_MASK(level)		(7 << CLIDR_CTYPE_SHIFT(level))
+#define CLIDR_CTYPE(clidr, level)	\
+	(((clidr) & CLIDR_CTYPE_MASK(level)) >> CLIDR_CTYPE_SHIFT(level))
+
+static inline enum cache_type get_cache_type(int level)
+{
+	unsigned int clidr;
+
+	if (level > MAX_CACHE_LEVEL)
+		return CACHE_TYPE_NOCACHE;
+	asm volatile ("mrc p15, 1, %0, c0, c0, 1" : "=r" (clidr));
+	return CLIDR_CTYPE(clidr, level);
+}
+
+/*
+ * NumSets, bits[27:13] - (Number of sets in cache) - 1
+ * Associativity, bits[12:3] - (Associativity of cache) - 1
+ * LineSize, bits[2:0] - (Log2(Number of words in cache line)) - 2
+ */
+#define CCSIDR_WRITE_THROUGH	BIT(31)
+#define CCSIDR_WRITE_BACK	BIT(30)
+#define CCSIDR_READ_ALLOCATE	BIT(29)
+#define CCSIDR_WRITE_ALLOCATE	BIT(28)
+#define CCSIDR_LINESIZE_MASK	0x7
+#define CCSIDR_ASSOCIAT_SHIFT	3
+#define CCSIDR_ASSOCIAT_MASK	0x3FF
+#define CCSIDR_NUMSETS_SHIFT	13
+#define CCSIDR_NUMSETS_MASK	0x7FF
+
+/*
+ * Which cache CCSIDR represents depends on CSSELR value
+ * Make sure no one else changes CSSELR during this
+ * smp_call_function_single prevents preemption for us
+ */
+static inline u32 get_ccsidr(u32 csselr)
+{
+	u32 ccsidr;
+
+	/* Put value into CSSELR */
+	asm volatile ("mcr p15, 2, %0, c0, c0, 0" : : "r" (csselr));
+	isb();
+	/* Read result out of CCSIDR */
+	asm volatile ("mrc p15, 1, %0, c0, c0, 0" : "=r" (ccsidr));
+
+	return ccsidr;
+}
+
+static void __ci_leaf_init(enum cache_type type, struct cacheinfo *this_leaf)
+{
+	bool is_instr_cache = type & CACHE_TYPE_INST;
+	u32 tmp = get_ccsidr((this_leaf->level - 1) << 1 | is_instr_cache);
+
+	this_leaf->type = type;
+	this_leaf->coherency_line_size =
+	    (1 << ((tmp & CCSIDR_LINESIZE_MASK) + 2)) * 4;
+	this_leaf->number_of_sets =
+	    ((tmp >> CCSIDR_NUMSETS_SHIFT) & CCSIDR_NUMSETS_MASK) + 1;
+	this_leaf->ways_of_associativity =
+	    ((tmp >> CCSIDR_ASSOCIAT_SHIFT) & CCSIDR_ASSOCIAT_MASK) + 1;
+	this_leaf->size = this_leaf->number_of_sets *
+	    this_leaf->coherency_line_size * this_leaf->ways_of_associativity;
+	this_leaf->attributes =
+		((tmp & CCSIDR_WRITE_THROUGH) ? CACHE_WRITE_THROUGH : 0) |
+		((tmp & CCSIDR_WRITE_BACK) ? CACHE_WRITE_BACK : 0) |
+		((tmp & CCSIDR_READ_ALLOCATE) ? CACHE_READ_ALLOCATE : 0) |
+		((tmp & CCSIDR_WRITE_ALLOCATE) ? CACHE_WRITE_ALLOCATE : 0);
+}
+
+#endif
+
+static void ci_leaf_init(struct cacheinfo *this_leaf,
+			 enum cache_type type, unsigned int level)
+{
+	this_leaf->level = level;
+	__ci_leaf_init(type, this_leaf);
+}
+
+int init_cache_level(unsigned int cpu)
+{
+	unsigned int ctype, level = 1, leaves = 0;
+	struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
+
+	if (!this_cpu_ci)
+		return -EINVAL;
+
+	do {
+		ctype = get_cache_type(level);
+		if (ctype == CACHE_TYPE_NOCACHE)
+			break;
+		/* Separate instruction and data caches */
+		leaves += (ctype == CACHE_TYPE_SEPARATE) ? 2 : 1;
+	} while (++level <= MAX_CACHE_LEVEL);
+
+	this_cpu_ci->num_levels = level - 1;
+	this_cpu_ci->num_leaves = leaves;
+
+	return 0;
+}
+
+int populate_cache_leaves(unsigned int cpu)
+{
+	unsigned int level, idx;
+	enum cache_type type;
+	struct cpu_cacheinfo *this_cpu_ci = get_cpu_cacheinfo(cpu);
+	struct cacheinfo *this_leaf = this_cpu_ci->info_list;
+
+	for (idx = 0, level = 1; level <= this_cpu_ci->num_levels &&
+	     idx < this_cpu_ci->num_leaves; idx++, level++) {
+		if (!this_leaf)
+			return -EINVAL;
+
+		type = get_cache_type(level);
+		if (type == CACHE_TYPE_SEPARATE) {
+			ci_leaf_init(this_leaf++, CACHE_TYPE_DATA, level);
+			ci_leaf_init(this_leaf++, CACHE_TYPE_INST, level);
+		} else {
+			ci_leaf_init(this_leaf++, type, level);
+		}
+	}
+	return 0;
+}
diff --git a/arch/arm/mm/Kconfig b/arch/arm/mm/Kconfig
index eda0dd0..fac8646 100644
--- a/arch/arm/mm/Kconfig
+++ b/arch/arm/mm/Kconfig
@@ -494,30 +494,42 @@  config CPU_PABRT_V7
 # The cache model
 config CPU_CACHE_V4
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_V4WT
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_V4WB
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_V6
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_V7
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_NOP
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_VIVT
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_VIPT
 	bool
+	select CPU_HAS_CACHE
 
 config CPU_CACHE_FA
 	bool
+	select CPU_HAS_CACHE
+
+config CPU_HAS_CACHE
+	bool
 
 if MMU
 # The copy-page model
@@ -845,6 +857,7 @@  config DMA_CACHE_RWFO
 
 config OUTER_CACHE
 	bool
+	select CPU_HAS_CACHE
 
 config OUTER_CACHE_SYNC
 	bool