>From the /sys/kernel/debug/ec/*/io file userspace can read and write EC byte
wise.
This patch introduces /sys/kernel/debug/ec/*/fields/* which exports all
fields of an Embedded Controller.
This has several advantages:
- Shows defined EC registers (not as dangerous as accessing arbitrary ones)
- Reads are done through the ACPI layer and may get accessed differently
than reading blank bytes (burst mode).
- People do not have to read the DSDT.dsl to map register names to EC fields
The patch has two unresolved isses:
- Unfortunately write access does not work for fields, acpica is missing this
capablity.
- All EC region fields (also SystemMemory, IO, etc.) are shown which may not
result in EC access. Ideally only the EC region fields are shown.
-> again acpica does not let us differ this.
Recent changes (v2):
- Return -EPERM if process has not CAP_SYS_RAWIO capability
- Do allow parallel access to EC files for stress testing ec.c driver
(instead of returning -EBUSY which was brought up akpm)
- Minor change in Kconfig description
Do not mention that EC is only accessed via ACPI interpreted code which
is not true for some native laptop driver hacks.
- Fix write_support file creation: mode and .write callback function, both
must be set (or removed) appropriately.
Signed-off-by: Thomas Renninger <trenn@xxxxxxx>
CC: mjg59@xxxxxxxxxxxxx
CC: platform-driver-x86@xxxxxxxxxxxxxxx
CC: linux-acpi@xxxxxxxxxxxxxxx
CC: astarikovskiy@xxxxxxx
CC: akpm@xxxxxxxxxxxxxxxxxxxx
CC: hmh@xxxxxxxxxx
---
Documentation/ABI/testing/debugfs-ec | 38 +++++++++-
drivers/acpi/Kconfig | 6 +-
drivers/acpi/ec_sys.c | 131 +++++++++++++++++++++++++++++++---
drivers/acpi/internal.h | 3 +
4 files changed, 163 insertions(+), 15 deletions(-)
diff --git a/Documentation/ABI/testing/debugfs-ec b/Documentation/ABI/testing/debugfs-ec
index 6546115..6bd5222 100644
--- a/Documentation/ABI/testing/debugfs-ec
+++ b/Documentation/ABI/testing/debugfs-ec
@@ -1,14 +1,21 @@
-What: /sys/kernel/debug/ec/*/{gpe,use_global_lock,io}
+What: /sys/kernel/debug/ec/*/{gpe,use_global_lock,io,field/*}
Date: July 2010
Contact: Thomas Renninger <trenn@xxxxxxx>
Description:
+files: gpe, use_global_lock
+---------------------------
+
General information like which GPE is assigned to the EC and whether
the global lock should get used.
-Knowing the EC GPE one can watch the amount of HW events related to
+Knowing the EC GPE, one can watch the amount of HW events related to
the EC here (XY -> GPE number from /sys/kernel/debug/ec/*/gpe):
/sys/firmware/acpi/interrupts/gpeXY
+
+file: io
+--------
+
The io file is binary and a userspace tool located here:
ftp://ftp.suse.com/pub/people/trenn/sources/ec/
should get used to read out the 256 Embedded Controller registers
@@ -18,3 +25,30 @@ CAUTION: Do not write to the Embedded Controller if you don't know
what you are doing! Rebooting afterwards also is a good idea.
This can influence the way your machine is cooled and fans may
not get switched on again after you did a wrong write.
+
+
+files: fields/*
+---------------
+
+Shows EC register fields as defined in the EmbeddedController object of the
+DSDT.
+Reads are done through the ACPI layer and may get accessed differently than
+reading blank bytes through the io file (e.g. with burst mode enabled).
+People can read the DSDT.dsl to map register names to EC fields, but with the
+names displayed this should not be necessary anymore.
+
+Typical EC region definition:
+ OperationRegion (ECRM, EmbeddedControl, 0x00, 0xFF)
+ Field (ECRM, ByteAcc, NoLock, Preserve)
+ {
+ S0PW, 32,
+ S0CT, 32,
+ Offset (0x80),
+ FNSW, 1,
+ ...
+ }
+
+Reading /sys/../ec/*/fields/SOPW will read the first 4 bytes of the EC and
+display it.
+Reading /sys/../ec/*/fields/FNSW will read one byte at offset 0x80 and mask out
+ 7 bits and return a bit value (FNSW is defined as a one bit field).
diff --git a/drivers/acpi/Kconfig b/drivers/acpi/Kconfig
index 08e0140..490eeef 100644
--- a/drivers/acpi/Kconfig
+++ b/drivers/acpi/Kconfig
@@ -116,9 +116,9 @@ config ACPI_EC_DEBUGFS
some seconds.
An Embedded Controller typically is available on laptops and reads
sensor values like battery state and temperature.
- The kernel accesses the EC through ACPI parsed code provided by BIOS
- tables. This option allows to access the EC directly without ACPI
- code being involved.
+ This option allows to access the EC directly without ACPI code being
+ involved (io file) or by calling the ACPI interpreter and access EC
+ ACPI field objects (field files, read-only).
Thus this option is a debug option that helps to write ACPI drivers
and can be used to identify ACPI code or EC firmware bugs.
diff --git a/drivers/acpi/ec_sys.c b/drivers/acpi/ec_sys.c
index 0e869b3..2be0066 100644
--- a/drivers/acpi/ec_sys.c
+++ b/drivers/acpi/ec_sys.c
@@ -6,11 +6,19 @@
* Thomas Renninger <trenn@xxxxxxx>
*
* This work is licensed under the terms of the GNU GPL, version 2.
+ *
+ * Ideally this would not be a separate driver, but hook into the
+ * .add function of the ec.c driver.
+ * But the ec.c driver must be compiled in and it's essential that
+ * this debug module can be loaded at runtime.
+ * The driver relies on the fact that an ACPI EC object can
+ * never vanish at runtime.
*/
#include <linux/kernel.h>
#include <linux/acpi.h>
#include <linux/debugfs.h>
+#include <linux/capability.h>
#include "internal.h"
MODULE_AUTHOR("Thomas Renninger <trenn@xxxxxxx>");
@@ -30,10 +38,11 @@ struct sysdev_class acpi_ec_sysdev_class = {
static struct dentry *acpi_ec_debugfs_dir;
-static int acpi_ec_open_io(struct inode *i, struct file *f)
+static int acpi_ec_open(struct inode *i, struct file *f)
{
f->private_data = i->i_private;
- return 0;
+ if (!capable(CAP_SYS_RAWIO))
+ return -EPERM;
}
static ssize_t acpi_ec_read_io(struct file *f, char __user *buf,
@@ -96,18 +105,107 @@ static ssize_t acpi_ec_write_io(struct file *f, const char __user *buf,
return count;
}
+#ifdef ACPICA_CAN_DO_FIELD_WRITES
+/* ACPICA currently does not provide a function to write to field
+ objects */
+static ssize_t acpi_ec_write_field(struct file *f, const char __user *buf,
+ size_t count, loff_t *off)
+{
+ struct acpi_ec *ec = f->private_data;
+ char name[5];
+ union acpi_object arg0 = { .type = ACPI_TYPE_INTEGER };
+ struct acpi_object_list arg_list = { 1, &arg0 };
+ acpi_status status;
+
+ if (*off > 0)
+ return -EINVAL;
+
+ strncpy(name, f->f_dentry->d_name.name, 5);
+
+ if (strict_strtoll(buf, 0, &arg0.integer.value))
+ return -EINVAL;
+
+ /* This does not work, only real methods can be evaluted
+ with an arguement */
+ status = acpi_evaluate_object(ec->handle, name, &arg_list, NULL);
+ if (ACPI_FAILURE(status))
+ return -EIO;
+ *off = count;
+ return count;
+}
+#endif
+
+static ssize_t acpi_ec_read_field(struct file *f, char __user *buf,
+ size_t count, loff_t *off)
+{
+ struct acpi_ec *ec = f->private_data;
+ char name[5];
+ unsigned long long value;
+ acpi_status status;
+
+ if (*off > 0)
+ return 0;
+ strncpy(name, f->f_dentry->d_name.name, 5);
+
+ status = acpi_evaluate_integer(ec->handle, name, NULL, &value);
+ if (ACPI_FAILURE(status))
+ return -EIO;
+
+ count = *off = sprintf(buf, "0x%llx\n", value);
+ if (!*off)
+ return -EINVAL;
+ return count;
+}
+
static struct file_operations acpi_ec_io_ops = {
- .owner = THIS_MODULE,
- .open = acpi_ec_open_io,
- .read = acpi_ec_read_io,
- .write = acpi_ec_write_io,
+ .owner = THIS_MODULE,
+ .open = acpi_ec_open,
+ .read = acpi_ec_read_io,
+};
+
+static struct file_operations acpi_ec_field_ops = {
+ .owner = THIS_MODULE,
+ .open = acpi_ec_open,
+ .llseek = no_llseek,
+ .read = acpi_ec_read_field,
};
+static acpi_status acpi_ec_find_fields(acpi_handle handle, u32 level,
+ void *context, void **return_value)
+{
+ char node_name[5];
+ struct acpi_buffer buffer = { sizeof(node_name), node_name };
+ struct acpi_ec *ec = context;
+ acpi_status status;
+ mode_t mode = 0400;
+
+#ifdef ACPICA_CAN_DO_FIELD_WRITES
+ if (write_support) {
+ mode = 0600;
+ acpi_ec_field_ops.write = acpi_ec_write_field;
+ }
+#endif
+
+ status = acpi_get_name(handle, ACPI_SINGLE_NAME, &buffer);
+ if (ACPI_SUCCESS(status)) {
+ if (!debugfs_create_file(node_name, mode, ec->field_dir,
+ ec, &acpi_ec_field_ops))
+ return AE_ERROR;
+ }
+ return AE_OK;
+}
+
int acpi_ec_add_debugfs(struct acpi_ec *ec, unsigned int ec_device_count)
{
- struct dentry *dev_dir;
+ struct dentry *dev_dir, *field_dir;
char name[64];
mode_t mode = 0400;
+ acpi_status status;
+
+ if (write_support) {
+ mode = 0600;
+ acpi_ec_io_ops.write = acpi_ec_write_io;
+ }
if (ec_device_count == 0) {
acpi_ec_debugfs_dir = debugfs_create_dir("ec", NULL);
@@ -129,13 +227,26 @@ int acpi_ec_add_debugfs(struct acpi_ec *ec, unsigned int ec_device_count)
(u32 *)&first_ec->global_lock))
goto error;
- if (write_support)
- mode = 0600;
if (!debugfs_create_file("io", mode, dev_dir, ec, &acpi_ec_io_ops))
goto error;
+ /*
+ * Find and register all fields. For now we do not differ
+ * EmbeddedController operation region fields because acpica does not
+ * allow us to do that. So some fields may result in system memory,
+ * io port,.. and not EC accesses.
+ */
+ field_dir = debugfs_create_dir("fields", dev_dir);
+ if (!field_dir)
+ goto error;
+ ec->field_dir = field_dir;
- return 0;
+ status = acpi_walk_namespace(ACPI_TYPE_LOCAL_REGION_FIELD, ec->handle,
+ 1, acpi_ec_find_fields, NULL, ec,
+ NULL);
+ if (ACPI_FAILURE(status))
+ goto error;
+ return 0;
error:
debugfs_remove_recursive(acpi_ec_debugfs_dir);
return -ENOMEM;
diff --git a/drivers/acpi/internal.h b/drivers/acpi/internal.h
index 8ae2726..14d1bab 100644
--- a/drivers/acpi/internal.h
+++ b/drivers/acpi/internal.h
@@ -64,6 +64,9 @@ struct acpi_ec {
struct transaction *curr;
spinlock_t curr_lock;
struct sys_device sysdev;
+#if defined CONFIG_ACPI_EC_DEBUGFS || defined CONFIG_ACPI_EC_DEBUGFS_MODULE
+ struct dentry *field_dir;
+#endif
};
extern struct acpi_ec *first_ec;
--
1.6.3
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
