[PATCH] MIPS: Make CP0 config registers readable via sysfs.

[PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Steven J. Hill]

From: "Steven J. Hill" <sjhill@mips.com>

Allow reading of CP0 config registers via sysfs for each core
in the system. The registers will show up in sysfs at the path:

   /sys/devices/system/cpu/cpuX/configX

Only CP0 config registers 0 through 7 are currently supported.

Signed-off-by: Steven J. Hill <sjhill@mips.com>
---
 arch/mips/kernel/Makefile |    1 +
 arch/mips/kernel/sysfs.c  |   93 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 94 insertions(+)
 create mode 100644 arch/mips/kernel/sysfs.c

diff --git a/arch/mips/kernel/Makefile b/arch/mips/kernel/Makefile
index e034ad6..e208a6b 100644
--- a/arch/mips/kernel/Makefile
+++ b/arch/mips/kernel/Makefile
@@ -98,6 +98,7 @@ obj-$(CONFIG_PERF_EVENTS)	+= perf_event.o
 obj-$(CONFIG_HW_PERF_EVENTS)	+= perf_event_mipsxx.o
 
 obj-$(CONFIG_JUMP_LABEL)	+= jump_label.o
+obj-y				+= sysfs.o
 
 ifeq ($(CONFIG_CPU_MIPS32), y)
 #
diff --git a/arch/mips/kernel/sysfs.c b/arch/mips/kernel/sysfs.c
new file mode 100644
index 0000000..a64f559
--- /dev/null
+++ b/arch/mips/kernel/sysfs.c
@@ -0,0 +1,93 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Copyright (C) 2012 MIPS Technologies, Inc.  All rights reserved.
+ */
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/cpu.h>
+#include <linux/percpu.h>
+
+#include <asm/page.h>
+
+
+#define __BUILD_CP0_SYSFS(reg)					\
+static DEFINE_PER_CPU(unsigned int, cpu_config##reg);		\
+static ssize_t show_config##reg(struct device *dev,		\
+		struct device_attribute *attr, char *buf)	\
+{								\
+	struct cpu *cpu = container_of(dev, struct cpu, dev);	\
+	int n = snprintf(buf, PAGE_SIZE-2, "%x\n",		\
+		per_cpu(cpu_config##reg, cpu->dev.id));		\
+	return n;						\
+}								\
+static DEVICE_ATTR(config##reg, 0444, show_config##reg, NULL);
+
+__BUILD_CP0_SYSFS(0)
+__BUILD_CP0_SYSFS(1)
+__BUILD_CP0_SYSFS(2)
+__BUILD_CP0_SYSFS(3)
+__BUILD_CP0_SYSFS(4)
+__BUILD_CP0_SYSFS(5)
+__BUILD_CP0_SYSFS(6)
+__BUILD_CP0_SYSFS(7)
+
+static void read_c0_registers(void *arg)
+{
+	struct device *dev = get_cpu_device(smp_processor_id());
+	struct cpu *cpu;
+	int ok;
+
+	if (dev != NULL) {
+		cpu = container_of(dev, struct cpu, dev);
+		per_cpu(cpu_config0, cpu->dev.id) = read_c0_config();
+		device_create_file(dev, &dev_attr_config0);
+		ok = per_cpu(cpu_config0, cpu->dev.id) & MIPS_CONF_M;
+	} else
+		return;
+
+	if (ok) {
+		per_cpu(cpu_config1, cpu->dev.id) = read_c0_config1();
+		device_create_file(dev, &dev_attr_config1);
+		ok = per_cpu(cpu_config1, cpu->dev.id) & MIPS_CONF_M;
+	}
+	if (ok) {
+		per_cpu(cpu_config2, cpu->dev.id) = read_c0_config2();
+		device_create_file(dev, &dev_attr_config2);
+		ok = per_cpu(cpu_config2, cpu->dev.id) & MIPS_CONF_M;
+	}
+	if (ok) {
+		per_cpu(cpu_config3, cpu->dev.id) = read_c0_config3();
+		device_create_file(dev, &dev_attr_config3);
+		ok = per_cpu(cpu_config3, cpu->dev.id) & MIPS_CONF_M;
+	}
+	if (ok) {
+		per_cpu(cpu_config4, cpu->dev.id) = read_c0_config4();
+		device_create_file(dev, &dev_attr_config4);
+		ok = per_cpu(cpu_config4, cpu->dev.id) & MIPS_CONF_M;
+	}
+	if (ok) {
+		per_cpu(cpu_config5, cpu->dev.id) = read_c0_config5();
+		device_create_file(dev, &dev_attr_config5);
+		ok = per_cpu(cpu_config5, cpu->dev.id) & MIPS_CONF_M;
+	}
+	if (ok) {
+		per_cpu(cpu_config6, cpu->dev.id) = read_c0_config6();
+		device_create_file(dev, &dev_attr_config6);
+		ok = per_cpu(cpu_config6, cpu->dev.id) & MIPS_CONF_M;
+	}
+	if (ok) {
+		per_cpu(cpu_config7, cpu->dev.id) = read_c0_config7();
+		device_create_file(dev, &dev_attr_config7);
+		ok = per_cpu(cpu_config7, cpu->dev.id) & MIPS_CONF_M;
+	}
+}
+
+static int __init mips_sysfs_registers(void)
+{
+	on_each_cpu(read_c0_registers, NULL, 1);
+	return 0;
+}
+late_initcall(mips_sysfs_registers);
-- 
1.7.9.5

Re: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Florian Fainelli]

Hello Steven,

Le 12/07/12 06:31, Steven J. Hill a écrit :

[snip]

> +
> +#define __BUILD_CP0_SYSFS(reg)					\
> +static DEFINE_PER_CPU(unsigned int, cpu_config##reg);		\
> +static ssize_t show_config##reg(struct device *dev,		\
> +		struct device_attribute *attr, char *buf)	\
> +{								\
> +	struct cpu *cpu = container_of(dev, struct cpu, dev);	\
> +	int n = snprintf(buf, PAGE_SIZE-2, "%x\n",		\
> +		per_cpu(cpu_config##reg, cpu->dev.id));		\
> +	return n;						\
> +}								\
> +static DEVICE_ATTR(config##reg, 0444, show_config##reg, NULL);
> +
> +__BUILD_CP0_SYSFS(0)
> +__BUILD_CP0_SYSFS(1)
> +__BUILD_CP0_SYSFS(2)
> +__BUILD_CP0_SYSFS(3)
> +__BUILD_CP0_SYSFS(4)
> +__BUILD_CP0_SYSFS(5)
> +__BUILD_CP0_SYSFS(6)
> +__BUILD_CP0_SYSFS(7)
> +
> +static void read_c0_registers(void *arg)
> +{
> +	struct device *dev = get_cpu_device(smp_processor_id());
> +	struct cpu *cpu;
> +	int ok;
> +
> +	if (dev != NULL) {
> +		cpu = container_of(dev, struct cpu, dev);
> +		per_cpu(cpu_config0, cpu->dev.id) = read_c0_config();
> +		device_create_file(dev, &dev_attr_config0);
> +		ok = per_cpu(cpu_config0, cpu->dev.id) & MIPS_CONF_M;
> +	} else

Is there any reason you are not using a macro here too?
--
Florian

Re: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Lars-Peter Clausen]

On 12/07/2012 06:31 AM, Steven J. Hill wrote:
> From: "Steven J. Hill" <sjhill@mips.com>
> 
> Allow reading of CP0 config registers via sysfs for each core
> in the system.

Maybe a stupid question, but why?

> The registers will show up in sysfs at the path:
> 
>    /sys/devices/system/cpu/cpuX/configX
> 
> Only CP0 config registers 0 through 7 are currently supported.

> 
> Signed-off-by: Steven J. Hill <sjhill@mips.com>
[...]

RE: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Hill, Steven]

Lars,

This patch was requested by our DSP/Codec group to help with selecting the best user-space codecs at runtime. Simply reading /proc/cpuinfo was insufficient. I posted this patch more for feedback and interest with minimal expectations that it would make it upstream. This patch will always be in our supported branches, but I will defer to everyone else on its worth for upstream.

-Steve

RE: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Hill, Steven]

By golly you're right! I can use a macro for the other code block. Thanks.

Re: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Florian Fainelli]

Le 12/12/12 17:44, Hill, Steven a écrit :
> Lars,
>
> This patch was requested by our DSP/Codec group to help with selecting the best user-space codecs at runtime. Simply reading /proc/cpuinfo was insufficient. I posted this patch more for feedback and interest with minimal expectations that it would make it upstream. This patch will always be in our supported branches, but I will defer to everyone else on its worth for upstream.

How do they actually make their choices? Is it just about advertising 
dsp vs. dspr2 to user-space?

Have you looked into modifying the ELF platform via set_elf_platform() 
to help these binaries with identify the underlying platform/kernel?
--
Florian

Re: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Lars-Peter Clausen]

On 12/12/2012 05:44 PM, Hill, Steven wrote:
> Lars,
> 
> This patch was requested by our DSP/Codec group to help with selecting the best user-space codecs at runtime. Simply reading /proc/cpuinfo was insufficient. I posted this patch more for feedback and interest with minimal expectations that it would make it upstream. This patch will always be in our supported branches, but I will defer to everyone else on its worth for upstream.
> 
> -Steve

Well if it is something that is useful it makes sense to upstream it,
especially if you are developing applications. Many people before you have
learned the hard way that stashing stuff away in their private branches was not
the best idea. If you are smart you are going to avoid that.

It may not be the best solution though to just dump all the cp register to
userspace. As Florian suggested there might be a smarter way to solve this.

- Lars

Re: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[David Daney]

On 12/12/2012 10:49 AM, Lars-Peter Clausen wrote:
> On 12/12/2012 05:44 PM, Hill, Steven wrote:
>> Lars,
>>
>> This patch was requested by our DSP/Codec group to help with selecting the best user-space codecs at runtime. Simply reading /proc/cpuinfo was insufficient. I posted this patch more for feedback and interest with minimal expectations that it would make it upstream. This patch will always be in our supported branches, but I will defer to everyone else on its worth for upstream.
>>
>> -Steve
>
> Well if it is something that is useful it makes sense to upstream it,
> especially if you are developing applications. Many people before you have
> learned the hard way that stashing stuff away in their private branches was not
> the best idea. If you are smart you are going to avoid that.
>
> It may not be the best solution though to just dump all the cp register to
> userspace. As Florian suggested there might be a smarter way to solve this.


If you want to have glibc's ld.so automatically select optimal libraries 
for a given platform, then the elf_platform is the way to do it.

However I don't think this is necessarily the case here.  elf_platform 
cannot easily handle a bunch of orthogonal capabilities.  So I am in 
favor of the principle of this patch.

There should be a per CPU file (probably in 
/sys/devices/system/cpu/cpuXX) with the CP0_Config values.

One question I have is:  Should it be a single file with one row for 
each implemented Config register, or one file per register?

David Daney

Re: [PATCH] MIPS: Make CP0 config registers readable via sysfs.

[Manuel Lauss]

On Wed, Dec 12, 2012 at 8:08 PM, David Daney <ddaney.cavm@gmail.com> wrote:
> On 12/12/2012 10:49 AM, Lars-Peter Clausen wrote:
>>
>> On 12/12/2012 05:44 PM, Hill, Steven wrote:
>>>
>>> Lars,
>>>
>>> This patch was requested by our DSP/Codec group to help with selecting
>>> the best user-space codecs at runtime. Simply reading /proc/cpuinfo was
>>> insufficient. I posted this patch more for feedback and interest with
>>> minimal expectations that it would make it upstream. This patch will always
>>> be in our supported branches, but I will defer to everyone else on its worth
>>> for upstream.
>>>
>>> -Steve
>>
>>
>> Well if it is something that is useful it makes sense to upstream it,
>> especially if you are developing applications. Many people before you have
>> learned the hard way that stashing stuff away in their private branches
>> was not
>> the best idea. If you are smart you are going to avoid that.
>>
>> It may not be the best solution though to just dump all the cp register to
>> userspace. As Florian suggested there might be a smarter way to solve
>> this.
>
>
>
> If you want to have glibc's ld.so automatically select optimal libraries for
> a given platform, then the elf_platform is the way to do it.
>
> However I don't think this is necessarily the case here.  elf_platform
> cannot easily handle a bunch of orthogonal capabilities.  So I am in favor
> of the principle of this patch.
>
> There should be a per CPU file (probably in /sys/devices/system/cpu/cpuXX)
> with the CP0_Config values.
>
> One question I have is:  Should it be a single file with one row for each
> implemented Config register, or one file per register?


I'm all for the one-file-per-register approach.  I find it easier to
just read the
right register instead of having to find out how many rows to read to get
the value I'm looking for.

Thanks!
        Manuel