Skip to contentSkip to conversation

Request driver inclusion - acer aspire one fan control

PATCH In review Peter Feuerer 7 participants 27 messages Jun 3, 2009 – Jun 19, 2009

Subsystems (2)
lore
Share & tools
lore

Peter Feuerer

author maintainer
#0 lore raw

PATCH 1/1 Request driver inclusion - acer aspire one fan control

Hi Len,

please include following driver.

thanks and best regards,
--peter

Acerhdf is a driver for Acer Aspire One netbooks. It allows to access the temperature sensor and to control the fan.

MAINTAINERS +0 -0
index cf4abdd..e9457fe 100644
@@ -222,6 +222,13 @@ L:	[email protected]

S: Maintained
F: drivers/net/acenic*

+ACER ASPIRE ONE TEMPERATURE AND FAN DRIVER
+P: Peter Feuerer
+M: [email protected]
+W: http://piie.net/?section=acerhdf
+S: Maintained
+F: drivers/platform/x86/acerhdf.c
+
ACER WMI LAPTOP EXTRAS
P: Carlos Corbacho
M: [email protected]

drivers/platform/x86/Kconfig +0 -0
index 284ebac..fe14dfd 100644
@@ -34,6 +34,23 @@ config ACER_WMI

If you have an ACPI-WMI compatible Acer/ Wistron laptop, say Y or M here.

+config ACERHDF
+ tristate "Acer Aspire One temperature and fan driver"
+ depends on THERMAL && THERMAL_HWMON
+ ---help---
+ This is a driver for Acer Aspire One netbooks. It allows to access
+ the temperature sensor and to control the fan.
+

+	  After loading this driver the BIOS is still in control of the fan.
+	  To let the kernel handle the fan, do:
+	  echo -n enabled > /sys/class/thermal/thermal_zone0/mode

+
+ For more information about this driver see
+ <http://piie.net/files/acerhdf_README.txt>
+
+ If you have an Acer Aspire One netbook, say Y or M
+ here.
+
config ASUS_LAPTOP
tristate "Asus Laptop Extras (EXPERIMENTAL)" depends on ACPI

drivers/platform/x86/Makefile +0 -0
index e40c7bd..641b8bf 100644
@@ -9,6 +9,7 @@ obj-$(CONFIG_COMPAL_LAPTOP)	+= compal-laptop.o

obj-$(CONFIG_DELL_LAPTOP) += dell-laptop.o
obj-$(CONFIG_DELL_WMI) += dell-wmi.o
obj-$(CONFIG_ACER_WMI) += acer-wmi.o
+obj-$(CONFIG_ACERHDF) += acerhdf.o
obj-$(CONFIG_HP_WMI) += hp-wmi.o
obj-$(CONFIG_TC1100_WMI) += tc1100-wmi.o
obj-$(CONFIG_SONY_LAPTOP) += sony-laptop.o

drivers/platform/x86/acerhdf.c new file +533 -0

533 lines not on this page · raw message

#1 lore raw

Hi,

Quoted 8 lines · Peter Feuerer

Hi Len,

please include following driver.

thanks and best regards,
--peter

Quoted 6 lines

Acerhdf is a driver for Acer Aspire One netbooks. It allows to access the temperature sensor and to control the fan.

By the way, your patch is line-wrapped and it won't apply here. See <Documentation/email-clients.txt> for more info.

Signature · 2 lines
Regards/Gruss,
Boris
author maintainer Reviewed-by Borislav Petkov Tested-by Borislav Petkov
#2 lore raw

Hi,

Quoted 2 lines · On Wed, 3 Jun 2009 14:14:52 +0200 Borislav Petkov <[email protected]>…

By the way, your patch is line-wrapped and it won't apply here. See <Documentation/email-clients.txt> for more info.

Seems like the email client I usually use is not able to send emails without the "format=flowed" option. Switched to Sylpheed, I hope it's working now.

kind regards,
peter

---

Acerhdf is a driver for Acer Aspire One netbooks. It allows to access the temperature sensor and to control the fan.

MAINTAINERS +7 -0
index cf4abdd..e9457fe 100644
@@ -222,6 +222,13 @@ L:	[email protected]
 S:	Maintained
 F:	drivers/net/acenic*
 
+ACER ASPIRE ONE TEMPERATURE AND FAN DRIVER
+P: Peter Feuerer
+M: [email protected]
+W: http://piie.net/?section=acerhdf
+S: Maintained
+F: drivers/platform/x86/acerhdf.c
+
 ACER WMI LAPTOP EXTRAS
 P:	Carlos Corbacho
 M:	[email protected]
drivers/platform/x86/Kconfig +17 -0
index 284ebac..fe14dfd 100644
@@ -34,6 +34,23 @@ config ACER_WMI
 	  If you have an ACPI-WMI compatible Acer/ Wistron laptop, say Y or M
 	  here.
 
+config ACERHDF
+	tristate "Acer Aspire One temperature and fan driver"
+	depends on THERMAL && THERMAL_HWMON
+	---help---
+	  This is a driver for Acer Aspire One netbooks. It allows to access
+	  the temperature sensor and to control the fan.
+
+	  After loading this driver the BIOS is still in control of the fan.
+	  To let the kernel handle the fan, do:
+	  echo -n enabled > /sys/class/thermal/thermal_zone0/mode
+
+	  For more information about this driver see
+	  <http://piie.net/files/acerhdf_README.txt>
+
+	  If you have an Acer Aspire One netbook, say Y or M
+	  here.
+
 config ASUS_LAPTOP
 	tristate "Asus Laptop Extras (EXPERIMENTAL)"
 	depends on ACPI
drivers/platform/x86/Makefile +1 -0
index e40c7bd..641b8bf 100644
@@ -9,6 +9,7 @@ obj-$(CONFIG_COMPAL_LAPTOP)	+= compal-laptop.o
 obj-$(CONFIG_DELL_LAPTOP)	+= dell-laptop.o
 obj-$(CONFIG_DELL_WMI)		+= dell-wmi.o
 obj-$(CONFIG_ACER_WMI)		+= acer-wmi.o
+obj-$(CONFIG_ACERHDF)		+= acerhdf.o
 obj-$(CONFIG_HP_WMI)		+= hp-wmi.o
 obj-$(CONFIG_TC1100_WMI)	+= tc1100-wmi.o
 obj-$(CONFIG_SONY_LAPTOP)	+= sony-laptop.o
drivers/platform/x86/acerhdf.c new file +608 -0

608 lines not on this page · raw message

#3 lore raw
Quoted 10 lines · On Wed, 3 Jun 2009 23:24:01 +0200 Peter Feuerer <[email protected]> wrote:

Acerhdf is a driver for Acer Aspire One netbooks. It allows to access the temperature sensor and to control the fan.

...

drivers/platform/x86/acerhdf.c line 47

Quoted 5 lines
drivers/platform/x86/acerhdf.c new file +5 -0
+/*
+ * if you want the module to be started in kernel mode,
+ * define the following:
+ */
+#undef START_IN_KERNEL_MODE

What does this mean?

afacit this functionality is already there at runtime via the module/boot parameter, so we don't need the compile-time knob?

Quoted 1 line

+#define VERSION "0.5.5"

That must be a record version number for an initial submission :)

Quoted 14 lines
drivers/platform/x86/acerhdf.c new file +14 -0
+/*
+ * According to the Atom N270 datasheet,
+ * (http://download.intel.com/design/processor/datashts/320032.pdf) the
+ * CPU's optimal operating limits denoted in junction temperature as
+ * measured by the on-die thermal monitor are within 0 <= Tj <= 90. So,
+ * assume 89__C is critical temperature.
+ */
+#define ACERHDF_TEMP_CRIT 89
+#define ACERHDF_FAN_AUTO 1
+#define ACERHDF_FAN_OFF 0
+
+/*
+ * No matter what value the user puts into the fanon variable, turn on the fan
+ * at 80 degree Celsius to prevent hardware damage

drivers/platform/x86/acerhdf.c line 74

Quoted 10 lines
drivers/platform/x86/acerhdf.c new file +10 -0
+ */
+#define ACERHDF_MAX_FANON 80
+
+/*
+ * Maximal interval between to temperature checks is 20 seconds, as the die
+ * can get hot really fast under heavy load
+ */
+#define ACERHDF_MAX_INTERVAL 20
+
+#define ACERHDF_ERROR -0xffff

Gad. I had to write a C program to check that. 0xffff0001. I wonder if the compiler treats this as a signed or unsigned thing.

Can this be simplified?

Quoted 13 lines
drivers/platform/x86/acerhdf.c new file +13 -0
+
+
+#ifdef START_IN_KERNEL_MODE
+static int kernelmode = 1;
+#else
+static int kernelmode;
+#endif
+
+static unsigned int interval = 10;
+static unsigned int fanon = 63;
+static unsigned int fanoff = 58;
+static unsigned int verbose;
+static unsigned int fanstate = ACERHDF_FAN_AUTO;

drivers/platform/x86/acerhdf.c line 98

Quoted 10 lines
drivers/platform/x86/acerhdf.c new file +10 -0
+static int disable_kernelmode;
+static int bios_version = -1;
+static char force_bios[16];
+static unsigned int prev_interval;
+struct thermal_zone_device *acerhdf_thz_dev;
+struct thermal_cooling_device *acerhdf_cool_dev;
+struct platform_device *acerhdf_device;
+
+module_param(kernelmode, uint, 0);
+MODULE_PARM_DESC(kernelmode, "Kernel mode on / off");

What's kernel mode? This should be explained in the code somewhere so I'm the last to ask.

drivers/platform/x86/acerhdf.c line 605

Quoted 9 lines

...

drivers/platform/x86/acerhdf.c new file +6 -0
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Peter Feuerer");
+MODULE_DESCRIPTION("Aspire One temperature and fan driver");
+MODULE_ALIAS("dmi:*:*Acer*:*:");
+MODULE_ALIAS("dmi:*:*Gateway*:*:");
+MODULE_ALIAS("dmi:*:*Packard Bell*:*:");

It's a nice-looking driver.

#4 lore raw

Hi,

Quoted 12 lines · Andrew Morton
Quoted 10 lines · On Wed, 3 Jun 2009 23:24:01 +0200 Peter Feuerer <[email protected]> wrote:

Acerhdf is a driver for Acer Aspire One netbooks. It allows to access the temperature sensor and to control the fan.

...

Quoted 7 lines

drivers/platform/x86/acerhdf.c line 47

Quoted 5 lines
drivers/platform/x86/acerhdf.c new file +5 -0
+/*
+ * if you want the module to be started in kernel mode,
+ * define the following:
+ */
+#undef START_IN_KERNEL_MODE

What does this mean?

when you boot the machine, the BIOS controlls the fan and it, noisy as it is, gets turned on at temperatures around 35-40°C. However, you can push that triggering point a bit higher when the machine is idle so that you don't get annoyed by the noisy fan. Actually, this is what this driver does.

The current design is that the driver gets loaded but still doesn't control the fan, as a precaution measure - we trust the BIOS (We didn't want to kill any user's machine through overheating.) If you turn on the START_IN_KERNEL_MODE, you circumvent that and the driver takes over the fan upon module load (I guess this is what you meant). Alternatively, you can do that over sysfs.

Quoted 6 lines

afacit this functionality is already there at runtime via the module/boot parameter, so we don't need the compile-time knob?

Quoted 1 line

+#define VERSION "0.5.5"

That must be a record version number for an initial submission :)

:)

Quoted 19 lines
Quoted 19 lines
drivers/platform/x86/acerhdf.c new file +19 -0
+/*
+ * According to the Atom N270 datasheet,
+ * (http://download.intel.com/design/processor/datashts/320032.pdf) the
+ * CPU's optimal operating limits denoted in junction temperature as
+ * measured by the on-die thermal monitor are within 0 <= Tj <= 90. So,
+ * assume 89__C is critical temperature.
+ */
+#define ACERHDF_TEMP_CRIT 89
+#define ACERHDF_FAN_AUTO 1
+#define ACERHDF_FAN_OFF 0
+
+/*
+ * No matter what value the user puts into the fanon variable, turn on the fan
+ * at 80 degree Celsius to prevent hardware damage
+ */
+#define ACERHDF_MAX_FANON 80
+
+/*
+ * Maximal interval between to temperature checks is 20 seconds, as the die
Quoted 8 lines

drivers/platform/x86/acerhdf.c line 74

Quoted 5 lines

+ * can get hot really fast under heavy load
+ */

drivers/platform/x86/acerhdf.c new file +3 -0
+#define ACERHDF_MAX_INTERVAL 20
+
+#define ACERHDF_ERROR -0xffff

Gad.  I had to write a C program to check that.  0xffff0001.  I wonder if the compiler treats this as a signed or unsigned thing.

it should be signed int.

Quoted 1 line

Can this be simplified?

This was supposed to be a random invalid value, I guess something like

#define ACERHDF_ERROR ~0

@Peter: Also, acerhdf_set_cur_state() needs a fix around

+ if (state == 0) {
+ /* turn fan off only if below fanoff temperature */
+ if ((cur_state == ACERHDF_FAN_AUTO) &&
+ (acerhdf_get_temp() < fanoff))

since acerhdf_get_temp() returns the temp _or_ and error. The cleaner thing should be if acerhdf_get_temp() returned the temp over a pointer arg just like the thermal callbacks do (e.g. acerhdf_get_ec_temp()) and you check the return value first and then interpret the temperature as valid.

Quoted 18 lines
Quoted 18 lines
drivers/platform/x86/acerhdf.c new file +18 -0
+
+
+#ifdef START_IN_KERNEL_MODE
+static int kernelmode = 1;
+#else
+static int kernelmode;
+#endif
+
+static unsigned int interval = 10;
+static unsigned int fanon = 63;
+static unsigned int fanoff = 58;
+static unsigned int verbose;
+static unsigned int fanstate = ACERHDF_FAN_AUTO;
+static int disable_kernelmode;
+static int bios_version = -1;
+static char force_bios[16];
+static unsigned int prev_interval;
+struct thermal_zone_device *acerhdf_thz_dev;
Quoted 8 lines

drivers/platform/x86/acerhdf.c line 98

Quoted 5 lines
drivers/platform/x86/acerhdf.c new file +5 -0
+struct thermal_cooling_device *acerhdf_cool_dev;
+struct platform_device *acerhdf_device;
+
+module_param(kernelmode, uint, 0);
+MODULE_PARM_DESC(kernelmode, "Kernel mode on / off");

What's kernel mode?  This should be explained in the code somewhere so I'm the last to ask.

see above.

Quoted 11 lines

drivers/platform/x86/acerhdf.c line 605

Quoted 9 lines

...

drivers/platform/x86/acerhdf.c new file +6 -0
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Peter Feuerer");
+MODULE_DESCRIPTION("Aspire One temperature and fan driver");
+MODULE_ALIAS("dmi:*:*Acer*:*:");
+MODULE_ALIAS("dmi:*:*Gateway*:*:");
+MODULE_ALIAS("dmi:*:*Packard Bell*:*:");

It's a nice-looking driver.

hope this is not irony :).

Signature · 2 lines
Regards/Gruss,
Boris
author maintainer
#5 lore raw

Hi,

Quoted 16 lines · On Thu, 4 Jun 2009 12:38:42 +0200 Borislav Petkov <[email protected]>…

Hi,

Quoted 5 lines · Andrew Morton
Quoted 1 line · On Wed, 3 Jun 2009 23:24:01 +0200 Peter Feuerer <[email protected]> wrote:

+#undef START_IN_KERNEL_MODE

What does this mean?

when you boot the machine, the BIOS controlls the fan and it, noisy as it is, gets turned on at temperatures around 35-40°C. However, you can push that triggering point a bit higher when the machine is idle so that you don't get annoyed by the noisy fan. Actually, this is what this driver does.

Quoted 6 lines

The current design is that the driver gets loaded but still doesn't control the fan, as a precaution measure - we trust the BIOS (We didn't want to kill any user's machine through overheating.) If you turn on the START_IN_KERNEL_MODE, you circumvent that and the driver takes over the fan upon module load (I guess this is what you meant). Alternatively, you can do that over sysfs.

I added a little description about this in my code. I hope it's clear for everybody who reads the code.

Quoted 9 lines
Quoted 7 lines

afacit this functionality is already there at runtime via the module/boot parameter, so we don't need the compile-time knob?

Quoted 1 line

+#define VERSION "0.5.5"

That must be a record version number for an initial submission :)

:)

Now it's already 0.5.6 ;)

Quoted 5 lines
Quoted 4 lines
Quoted 1 line

+#define ACERHDF_ERROR -0xffff

Gad.  I had to write a C program to check that.  0xffff0001.  I wonder if the compiler treats this as a signed or unsigned thing.

Quoted 7 lines

it should be signed int.

Quoted 1 line

Can this be simplified?

This was supposed to be a random invalid value, I guess something like

#define ACERHDF_ERROR ~0

Changed it (see code).

Quoted 8 lines

@Peter: Also, acerhdf_set_cur_state() needs a fix around

(quoted diff) +4 -0
+       if (state == 0) {
+               /* turn fan off only if below fanoff temperature */
+               if ((cur_state == ACERHDF_FAN_AUTO) &&
+                               (acerhdf_get_temp() < fanoff))
Quoted 5 lines

since acerhdf_get_temp() returns the temp _or_ and error. The cleaner thing should be if acerhdf_get_temp() returned the temp over a pointer arg just like the thermal callbacks do (e.g. acerhdf_get_ec_temp()) and you check the return value first and then interpret the temperature as valid.

Yes that's right, this piece of code was missing.

Quoted 3 lines
Quoted 1 line

It's a nice-looking driver.

hope this is not irony :).

Thanks Andrew!

Signature · 10 lines
kind regards,
 peter

---
Acerhdf is a driver for Acer Aspire One netbooks. It allows to access
the temperature sensor and to control the fan.

Signed-off-by: Peter Feuerer <[email protected]>
Reviewed-by: Borislav Petkov <[email protected]>
Tested-by: Borislav Petkov <[email protected]>
MAINTAINERS +7 -0
index cf4abdd..e9457fe 100644
@@ -222,6 +222,13 @@ L:	[email protected]
 S:	Maintained
 F:	drivers/net/acenic*
 
+ACER ASPIRE ONE TEMPERATURE AND FAN DRIVER
+P: Peter Feuerer
+M: [email protected]
+W: http://piie.net/?section=acerhdf
+S: Maintained
+F: drivers/platform/x86/acerhdf.c
+
 ACER WMI LAPTOP EXTRAS
 P:	Carlos Corbacho
 M:	[email protected]
drivers/platform/x86/Kconfig +17 -0
index 284ebac..fe14dfd 100644
@@ -34,6 +34,23 @@ config ACER_WMI
 	  If you have an ACPI-WMI compatible Acer/ Wistron laptop, say Y or M
 	  here.
 
+config ACERHDF
+	tristate "Acer Aspire One temperature and fan driver"
+	depends on THERMAL && THERMAL_HWMON
+	---help---
+	  This is a driver for Acer Aspire One netbooks. It allows to access
+	  the temperature sensor and to control the fan.
+
+	  After loading this driver the BIOS is still in control of the fan.
+	  To let the kernel handle the fan, do:
+	  echo -n enabled > /sys/class/thermal/thermal_zone0/mode
+
+	  For more information about this driver see
+	  <http://piie.net/files/acerhdf_README.txt>
+
+	  If you have an Acer Aspire One netbook, say Y or M
+	  here.
+
 config ASUS_LAPTOP
 	tristate "Asus Laptop Extras (EXPERIMENTAL)"
 	depends on ACPI
drivers/platform/x86/Makefile +1 -0
index e40c7bd..641b8bf 100644
@@ -9,6 +9,7 @@ obj-$(CONFIG_COMPAL_LAPTOP)	+= compal-laptop.o
 obj-$(CONFIG_DELL_LAPTOP)	+= dell-laptop.o
 obj-$(CONFIG_DELL_WMI)		+= dell-wmi.o
 obj-$(CONFIG_ACER_WMI)		+= acer-wmi.o
+obj-$(CONFIG_ACERHDF)		+= acerhdf.o
 obj-$(CONFIG_HP_WMI)		+= hp-wmi.o
 obj-$(CONFIG_TC1100_WMI)	+= tc1100-wmi.o
 obj-$(CONFIG_SONY_LAPTOP)	+= sony-laptop.o
drivers/platform/x86/acerhdf.c new file +628 -0

628 lines not on this page · raw message

Reviewed-by Tested-by
#6 lore raw

Hi,

tested and working on Aspire One 110L, BIOS v0.3109 (oldest), 2.6.30-rc8, including resume (short testing only, though).

It's a nice thing to be able to replace the less efficient monitoring shell script stop-gap (many thanks to the people that provided this very useful interim infrastructure, though!)

The current patch seems fine to me for inclusion,
and given Peter's prior performance I'm highly confident that a couple rather minor review comments will get addressed, too.

#7 lore raw

[PATCH/RFC] Acer Aspire One fan control resume fix, improvements

In reply to Peter Feuerer

Hello all,

tried version 0.5.6, didn't (quite!) like it. ;)
Reason being that when suspending, every couple times there was a fan state check error on resume, leading to the annoying issue of kernel mode fan control getting disabled (via a safety measure).

ChangeLog:
- reduce maximum temperature check interval, 20 seconds seems a bit much (we have to take into account _external_ heat sources, too!)
- configure BIOS mode during suspend downtime, since our driver code is having a day off (and go back to previous setting after resume)
- update fan state variable during resume, to reflect new state after powering up
- optimization: add bios_settings pointer to current bios's settings, avoids array indirection
- change cmd_off, cmd_auto to fancmd[2] for streamlined code
- several improved log messages
- reverse fan state printing (every value that doesn't indicate "OFF" should definitely lead towards an "AUTO" fan setting!!!!)
[some functional check was unsafe in this respect, too]
- use LONG_MAX instead of open-coded 0x7fffffffl

Suspend tested multiple times (on 2.6.30-rc8), checkpatch.pl:ed. Version 0.5.7 was internal release only.
This patch is intended for Peter mainly, he should decide what to do with that content - better don't commit it yet.

drivers/platform/x86/acerhdf.c renamed from drivers/platform/x86/acerhdf.c.patched_orig +82 -55
@@ -13,6 +13,7 @@
  *               - Carlos Corbacho  cathectic (at) gmail.com
  *  o lkml       - Matthew Garrett
  *               - Borislav Petkov
+ *               - Andreas Mohr
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -52,7 +53,7 @@
  */
 #undef START_IN_KERNEL_MODE
 
-#define VERSION "0.5.6"
+#define VERSION "0.5.8"
 
 /*
  * According to the Atom N270 datasheet,
@@ -62,8 +63,8 @@
  * assume 89°C is critical temperature.
  */
 #define ACERHDF_TEMP_CRIT 89
-#define ACERHDF_FAN_AUTO 1
 #define ACERHDF_FAN_OFF 0
+#define ACERHDF_FAN_AUTO 1
 
 /*
  * No matter what value the user puts into the fanon variable, turn on the fan
@@ -72,17 +73,18 @@
 #define ACERHDF_MAX_FANON 80
 
 /*
- * Maximal interval between two temperature checks is 20 seconds, as the die
- * can get hot really fast under heavy load
+ * Maximum interval between two temperature checks is 15 seconds, as the die
+ * can get hot really fast under heavy load (plus we shouldn't forget about
+ * possible impact of _external_ aggressive sources such as heaters, sun etc.)
  */
-#define ACERHDF_MAX_INTERVAL 20
+#define ACERHDF_MAX_INTERVAL 15
 
 /*
  * As temperatures can be negative, zero or positive, the value indicating
  * an error must be somewhere beyond valid temperature values.
- * 0x7fffffffl - highest possible positive long value should do the job.
+ * LONG_MAX (highest possible positive long value) should do the job.
  */
-#define ACERHDF_ERROR 0x7fffffffl
+#define ACERHDF_ERROR LONG_MAX
 
 
 #ifdef START_IN_KERNEL_MODE
@@ -97,7 +99,7 @@
 static unsigned int verbose;
 static unsigned int fanstate = ACERHDF_FAN_AUTO;
 static int disable_kernelmode;
-static int bios_version = -1;
+static int pre_suspend_kernelmode;
 static char force_bios[16];
 static unsigned int prev_interval;
 struct thermal_zone_device *acerhdf_thz_dev;
@@ -123,25 +125,26 @@
 	const char *version;
 	unsigned char fanreg;
 	unsigned char tempreg;
-	unsigned char cmd_off;
-	unsigned char cmd_auto;
+	unsigned char fancmd[2]; /* fan off and auto commands */
 };
 
 /* Register addresses and values for different BIOS versions */
-static const struct bios_settings_t bios_settings[] = {
-	{"Acer", "v0.3109", 0x55, 0x58, 0x1f, 0x00},
-	{"Acer", "v0.3114", 0x55, 0x58, 0x1f, 0x00},
-	{"Acer", "v0.3301", 0x55, 0x58, 0xaf, 0x00},
-	{"Acer", "v0.3304", 0x55, 0x58, 0xaf, 0x00},
-	{"Acer", "v0.3305", 0x55, 0x58, 0xaf, 0x00},
-	{"Acer", "v0.3308", 0x55, 0x58, 0x21, 0x00},
-	{"Acer", "v0.3309", 0x55, 0x58, 0x21, 0x00},
-	{"Acer", "v0.3310", 0x55, 0x58, 0x21, 0x00},
-	{"Gateway", "v0.3103", 0x55, 0x58, 0x21, 0x00},
-	{"Packard Bell", "v0.3105", 0x55, 0x58, 0x21, 0x00},
-	{"", 0, 0, 0, 0, 0}
+static const struct bios_settings_t bios_settings_table[] = {
+	{"Acer", "v0.3109", 0x55, 0x58, {0x1f, 0x00} },
+	{"Acer", "v0.3114", 0x55, 0x58, {0x1f, 0x00} },
+	{"Acer", "v0.3301", 0x55, 0x58, {0xaf, 0x00} },
+	{"Acer", "v0.3304", 0x55, 0x58, {0xaf, 0x00} },
+	{"Acer", "v0.3305", 0x55, 0x58, {0xaf, 0x00} },
+	{"Acer", "v0.3308", 0x55, 0x58, {0x21, 0x00} },
+	{"Acer", "v0.3309", 0x55, 0x58, {0x21, 0x00} },
+	{"Acer", "v0.3310", 0x55, 0x58, {0x21, 0x00} },
+	{"Gateway", "v0.3103", 0x55, 0x58, {0x21, 0x00} },
+	{"Packard Bell", "v0.3105", 0x55, 0x58, {0x21, 0x00} },
+	{"", 0, 0, 0, {0, 0} }
 };
 
+static const struct bios_settings_t *bios_settings __read_mostly;
+
 
 /* acer ec functions */
 static int acerhdf_get_temp(void)
@@ -149,7 +152,7 @@
 	u8 temp;
 
 	/* read temperature */
-	if (!ec_read(bios_settings[bios_version].tempreg, &temp)) {
+	if (!ec_read(bios_settings->tempreg, &temp)) {
 		if (verbose)
 			pr_notice("temp %d\n", temp);
 		return temp;
@@ -161,8 +164,8 @@
 {
 	u8 fan;
 
-	if (!ec_read(bios_settings[bios_version].fanreg, &fan))
-		return (fan == bios_settings[bios_version].cmd_off) ?
+	if (!ec_read(bios_settings->fanreg, &fan))
+		return (fan == bios_settings->fancmd[ACERHDF_FAN_OFF]) ?
 			ACERHDF_FAN_OFF : ACERHDF_FAN_AUTO;
 
 	return ACERHDF_ERROR;
@@ -173,19 +176,19 @@
 	unsigned char cmd;
 
 	if (verbose)
-		pr_notice("fan %s\n", (state == ACERHDF_FAN_AUTO) ?
-				"ON" : "OFF");
+		pr_notice("fan %s\n", (state == ACERHDF_FAN_OFF) ?
+				"OFF" : "ON");
 
-	if (state == ACERHDF_FAN_AUTO) {
-		cmd = bios_settings[bios_version].cmd_auto;
-		fanstate = ACERHDF_FAN_AUTO;
-	} else {
-		cmd = bios_settings[bios_version].cmd_off;
-		fanstate = ACERHDF_FAN_OFF;
+	if ((state != ACERHDF_FAN_OFF) && (state != ACERHDF_FAN_AUTO)) {
+		pr_err("invalid fan state %d requested, setting to auto!\n",
+			state);
+		state = ACERHDF_FAN_AUTO;
 	}
 
-	ec_write(bios_settings[bios_version].fanreg, cmd);
+	cmd = bios_settings->fancmd[state];
+	fanstate = state;
 
+	ec_write(bios_settings->fanreg, cmd);
 }
 
 /* helpers */
@@ -253,6 +256,18 @@
 	return 0;
 }
 
+/* provide one central function to set disable_kernelmode
+ * (always set ACERHDF_FAN_AUTO, too!) */
+static inline void acerhdf_revert_to_bios_mode(void)
+{
+	acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
+	/*
+	 * let the thermal layer disable kernel mode. This ensures that
+	 * the thermal layer doesn't switch off the fan again
+	 */
+	disable_kernelmode = 1;
+}
+
 /*  current operation mode - enabled / disabled */
 static int acerhdf_get_mode(struct thermal_zone_device *thermal,
 		enum thermal_device_mode *mode)
@@ -279,13 +294,8 @@
 		pr_notice("kernel mode OFF\n");
 		thermal->polling_delay = 0;
 		thermal_zone_device_update(thermal);
-		acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
 
-		/*
-		 * let the thermal layer disable kernel mode. This ensures that
-		 * the thermal layer doesn't switch off the fan again
-		 */
-		disable_kernelmode = 1;
+		acerhdf_revert_to_bios_mode();
 	} else {
 		kernelmode = 1;
 		pr_notice("kernel mode ON\n");
@@ -394,21 +404,19 @@
 	 */
 	if (cur_state != fanstate) {
 		pr_err("failed reading fan state, "
-				"disabling kernelmode.\n");
+				"falling back to default BIOS handling.\n");
 		pr_err("read state: %d expected state: %d\n",
 				cur_state, fanstate);
 
-		acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
-		disable_kernelmode = 1;
+		acerhdf_revert_to_bios_mode();
 		return -EINVAL;
 	}
 	/* same with temperature */
 	if (cur_temp == ACERHDF_ERROR) {
 		pr_err("failed reading temperature, "
-				"disabling kernelmode.\n");
+				"falling back to default BIOS handling.\n");
 
-		acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
-		disable_kernelmode = 1;
+		acerhdf_revert_to_bios_mode();
 		return -EINVAL;
 	}
 
@@ -437,11 +445,21 @@
 		pm_message_t state)
 {
 	/*
-	 * in kernelmode turn on fan, because the aspire one awakes with
-	 * spinning fan
+	 * always revert to BIOS mode during suspend/resume activities:
+	 * a) during suspend our driver is inactive, thus if there's
+	 *    anything to be done fan-wise, it's the BIOS's job...
+	 * b) Aspire awakes with spinning fan in BIOS mode,
+	 *    thus we better do the same (behaviour is preserved if we use BIOS)
 	 */
-	if (kernelmode)
+
+	/* remember previous setting */
+	pre_suspend_kernelmode = kernelmode;
+
+	if (kernelmode) {
 		acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
+		kernelmode = 0;
+	}
+
 	if (verbose)
 		pr_notice("going suspend\n");
 	return 0;
@@ -449,6 +467,14 @@
 
 static int acerhdf_resume(struct platform_device *device)
 {
+	kernelmode = pre_suspend_kernelmode;
+
+	/* update our fanstate variable to the possibly different
+	 * post-resume fan state
+	 * (to prevent a safety check from failing)
+	 */
+	fanstate = acerhdf_get_fanstate();
+
 	if (verbose)
 		pr_notice("resuming\n");
 	return 0;
@@ -526,15 +552,16 @@
 
 
 	/* search BIOS version and BIOS vendor in BIOS settings table */
-	for (i = 0; bios_settings[i].version[0]; ++i) {
-		if (!strcmp(bios_settings[i].vendor, vendor) &&
-				!strcmp(bios_settings[i].version, version)) {
-			bios_version = i;
+	for (i = 0; bios_settings_table[i].version[0]; ++i) {
+		if (!strcmp(bios_settings_table[i].vendor, vendor) &&
+		    !strcmp(bios_settings_table[i].version, version)) {
+			bios_settings = &bios_settings_table[i];
 			break;
 		}
 	}
-	if (bios_version == -1) {
-		pr_err("cannot find BIOS version\n");
+	if (!bios_settings) {
+		pr_err("unknown (unsupported) BIOS version %s/%s, "
+			"please report, aborting!\n", vendor, version);
 		return ACERHDF_ERROR;
 	}
 	return 0;
#8 lore raw

Re: [PATCH/RFC] Acer Aspire One fan control resume fix, improvements

Hi,

Quoted 7 lines · Andreas Mohr

Hello all,

tried version 0.5.6, didn't (quite!) like it. ;)
Reason being that when suspending, every couple times there was a fan state check error on resume, leading to the annoying issue of kernel mode fan control getting disabled (via a safety measure).

Quoted 13 lines

ChangeLog:
- reduce maximum temperature check interval, 20 seconds seems a bit much (we have to take into account _external_ heat sources, too!)
- configure BIOS mode during suspend downtime, since our driver code is having a day off (and go back to previous setting after resume)
- update fan state variable during resume, to reflect new state after powering up
- optimization: add bios_settings pointer to current bios's settings, avoids array indirection
- change cmd_off, cmd_auto to fancmd[2] for streamlined code
- several improved log messages
- reverse fan state printing (every value that doesn't indicate "OFF" should definitely lead towards an "AUTO" fan setting!!!!)

good catch!

Quoted 64 lines

 [some functional check was unsafe in this respect, too]
- use LONG_MAX instead of open-coded 0x7fffffffl

Suspend tested multiple times (on 2.6.30-rc8), checkpatch.pl:ed. Version 0.5.7 was internal release only.
This patch is intended for Peter mainly, he should decide what to do with that content - better don't commit it yet.

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +9 -7
@@ -13,6 +13,7 @@
  *               - Carlos Corbacho  cathectic (at) gmail.com
  *  o lkml       - Matthew Garrett
  *               - Borislav Petkov
+ *               - Andreas Mohr
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -52,7 +53,7 @@
  */
  #undef START_IN_KERNEL_MODE
 
-#define VERSION "0.5.6"
+#define VERSION "0.5.8"
 
  /*
  * According to the Atom N270 datasheet,
@@ -62,8 +63,8 @@
  * assume 89°C is critical temperature.
  */
  #define ACERHDF_TEMP_CRIT 89
-#define ACERHDF_FAN_AUTO 1
  #define ACERHDF_FAN_OFF 0
+#define ACERHDF_FAN_AUTO 1
 
  /*
  * No matter what value the user puts into the fanon variable, turn on the fan
@@ -72,17 +73,18 @@
  #define ACERHDF_MAX_FANON 80
 
  /*
- * Maximal interval between two temperature checks is 20 seconds, as the die
- * can get hot really fast under heavy load
+ * Maximum interval between two temperature checks is 15 seconds, as the die
+ * can get hot really fast under heavy load (plus we shouldn't forget about
+ * possible impact of _external_ aggressive sources such as heaters, sun etc.)
  */
-#define ACERHDF_MAX_INTERVAL 20
+#define ACERHDF_MAX_INTERVAL 15
 
  /*
  * As temperatures can be negative, zero or positive, the value indicating
  * an error must be somewhere beyond valid temperature values.
- * 0x7fffffffl - highest possible positive long value should do the job.
+ * LONG_MAX (highest possible positive long value) should do the job.
  */
-#define ACERHDF_ERROR 0x7fffffffl
+#define ACERHDF_ERROR LONG_MAX
 
 
  #ifdef START_IN_KERNEL_MODE

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 line 102

Quoted 5 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +1 -1
@@ -97,7 +99,7 @@
  static unsigned int verbose;
  static unsigned int fanstate = ACERHDF_FAN_AUTO;
  static int disable_kernelmode;
-static int bios_version = -1;
+static int pre_suspend_kernelmode;

ok, this variable is unnecessary:

it is enough to do

if (kernelmode)

	acerhdf_change_fanstate(ACERHDF_FAN_AUTO);

on suspend and since the machine starts with the BIOS in control of the fan, the acerhdf_set_cur_state() thermal callback will figure out what to do so you don't need to explicitly save the kernelmode setting before suspend and restore it on resume.

Quoted 179 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +40 -32
  static char force_bios[16];
  static unsigned int prev_interval;
  struct thermal_zone_device *acerhdf_thz_dev;
@@ -123,25 +125,26 @@
        const char *version;
        unsigned char fanreg;
        unsigned char tempreg;
-       unsigned char cmd_off;
-       unsigned char cmd_auto;
+       unsigned char fancmd[2]; /* fan off and auto commands */
  };
 
  /* Register addresses and values for different BIOS versions */
-static const struct bios_settings_t bios_settings[] = {
-       {"Acer", "v0.3109", 0x55, 0x58, 0x1f, 0x00},
-       {"Acer", "v0.3114", 0x55, 0x58, 0x1f, 0x00},
-       {"Acer", "v0.3301", 0x55, 0x58, 0xaf, 0x00},
-       {"Acer", "v0.3304", 0x55, 0x58, 0xaf, 0x00},
-       {"Acer", "v0.3305", 0x55, 0x58, 0xaf, 0x00},
-       {"Acer", "v0.3308", 0x55, 0x58, 0x21, 0x00},
-       {"Acer", "v0.3309", 0x55, 0x58, 0x21, 0x00},
-       {"Acer", "v0.3310", 0x55, 0x58, 0x21, 0x00},
-       {"Gateway", "v0.3103", 0x55, 0x58, 0x21, 0x00},
-       {"Packard Bell", "v0.3105", 0x55, 0x58, 0x21, 0x00},
-       {"", 0, 0, 0, 0, 0}
+static const struct bios_settings_t bios_settings_table[] = {
+       {"Acer", "v0.3109", 0x55, 0x58, {0x1f, 0x00} },
+       {"Acer", "v0.3114", 0x55, 0x58, {0x1f, 0x00} },
+       {"Acer", "v0.3301", 0x55, 0x58, {0xaf, 0x00} },
+       {"Acer", "v0.3304", 0x55, 0x58, {0xaf, 0x00} },
+       {"Acer", "v0.3305", 0x55, 0x58, {0xaf, 0x00} },
+       {"Acer", "v0.3308", 0x55, 0x58, {0x21, 0x00} },
+       {"Acer", "v0.3309", 0x55, 0x58, {0x21, 0x00} },
+       {"Acer", "v0.3310", 0x55, 0x58, {0x21, 0x00} },
+       {"Gateway", "v0.3103", 0x55, 0x58, {0x21, 0x00} },
+       {"Packard Bell", "v0.3105", 0x55, 0x58, {0x21, 0x00} },
+       {"", 0, 0, 0, {0, 0} }
  };
 
+static const struct bios_settings_t *bios_settings __read_mostly;
+
 
  /* acer ec functions */
  static int acerhdf_get_temp(void)
@@ -149,7 +152,7 @@
        u8 temp;
 
        /* read temperature */
-       if (!ec_read(bios_settings[bios_version].tempreg, &temp)) {
+       if (!ec_read(bios_settings->tempreg, &temp)) {
                if (verbose)
                        pr_notice("temp %d\n", temp);
                return temp;
@@ -161,8 +164,8 @@
  {
        u8 fan;
 
-       if (!ec_read(bios_settings[bios_version].fanreg, &fan))
-               return (fan == bios_settings[bios_version].cmd_off) ?
+       if (!ec_read(bios_settings->fanreg, &fan))
+               return (fan == bios_settings->fancmd[ACERHDF_FAN_OFF]) ?
                        ACERHDF_FAN_OFF : ACERHDF_FAN_AUTO;
 
        return ACERHDF_ERROR;
@@ -173,19 +176,19 @@
        unsigned char cmd;
 
        if (verbose)
-               pr_notice("fan %s\n", (state == ACERHDF_FAN_AUTO) ?
-                               "ON" : "OFF");
+               pr_notice("fan %s\n", (state == ACERHDF_FAN_OFF) ?
+                               "OFF" : "ON");
 
-       if (state == ACERHDF_FAN_AUTO) {
-               cmd = bios_settings[bios_version].cmd_auto;
-               fanstate = ACERHDF_FAN_AUTO;
-       } else {
-               cmd = bios_settings[bios_version].cmd_off;
-               fanstate = ACERHDF_FAN_OFF;
+       if ((state != ACERHDF_FAN_OFF) && (state != ACERHDF_FAN_AUTO)) {
+               pr_err("invalid fan state %d requested, setting to auto!\n",
+                       state);
+               state = ACERHDF_FAN_AUTO;
        }
 
-       ec_write(bios_settings[bios_version].fanreg, cmd);
+       cmd = bios_settings->fancmd[state];
+       fanstate = state;
 
+       ec_write(bios_settings->fanreg, cmd);
  }
 
  /* helpers */
@@ -253,6 +256,18 @@
        return 0;
  }
 
+/* provide one central function to set disable_kernelmode
+ * (always set ACERHDF_FAN_AUTO, too!) */
+static inline void acerhdf_revert_to_bios_mode(void)
+{
+       acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
+       /*
+        * let the thermal layer disable kernel mode. This ensures that
+        * the thermal layer doesn't switch off the fan again
+        */
+       disable_kernelmode = 1;
+}
+
  /*  current operation mode - enabled / disabled */
  static int acerhdf_get_mode(struct thermal_zone_device *thermal,
                enum thermal_device_mode *mode)
@@ -279,13 +294,8 @@
                pr_notice("kernel mode OFF\n");
                thermal->polling_delay = 0;
                thermal_zone_device_update(thermal);
-               acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
 
-               /*
-                * let the thermal layer disable kernel mode. This ensures that
-                * the thermal layer doesn't switch off the fan again
-                */
-               disable_kernelmode = 1;
+               acerhdf_revert_to_bios_mode();
        } else {
                kernelmode = 1;
                pr_notice("kernel mode ON\n");
@@ -394,21 +404,19 @@
         */
        if (cur_state != fanstate) {

               pr_err("failed reading fan state, "
-                               "disabling kernelmode.\n");
+                               "falling back to default BIOS handling.\n"); pr_err("read state: %d expected state: %d\n", cur_state, fanstate);

-               acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
-               disable_kernelmode = 1;
+               acerhdf_revert_to_bios_mode();
               return -EINVAL;
       }
       /* same with temperature */
       if (cur_temp == ACERHDF_ERROR) {
               pr_err("failed reading temperature, "
-                               "disabling kernelmode.\n");
+                               "falling back to default BIOS handling.\n");

-               acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
-               disable_kernelmode = 1;
+               acerhdf_revert_to_bios_mode(); return -EINVAL;
       }

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +13 -3
@@ -437,11 +445,21 @@
                pm_message_t state)
  {
        /*
-        * in kernelmode turn on fan, because the aspire one awakes with
-        * spinning fan
+        * always revert to BIOS mode during suspend/resume activities:
+        * a) during suspend our driver is inactive, thus if there's
+        *    anything to be done fan-wise, it's the BIOS's job...
+        * b) Aspire awakes with spinning fan in BIOS mode,
+        *    thus we better do the same (behaviour is preserved if we use BIOS)
         */
-       if (kernelmode)
+
+       /* remember previous setting */
+       pre_suspend_kernelmode = kernelmode;
+
+       if (kernelmode) {
                acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
+               kernelmode = 0;
+       }
+
        if (verbose)
                pr_notice("going suspend\n");
        return 0;

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 line 475

Quoted 9 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +6 -0
@@ -449,6 +467,14 @@
 
  static int acerhdf_resume(struct platform_device *device)
  {
+       kernelmode = pre_suspend_kernelmode;
+
+       /* update our fanstate variable to the possibly different
+        * post-resume fan state
+        * (to prevent a safety check from failing)
+        */

please use kernel-style comments:

/*
* comment text goes here
*/

Quoted 29 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +9 -6
+       fanstate = acerhdf_get_fanstate();
+
        if (verbose)
                pr_notice("resuming\n");
        return 0;
@@ -526,15 +552,16 @@
 
 
        /* search BIOS version and BIOS vendor in BIOS settings table */
-       for (i = 0; bios_settings[i].version[0]; ++i) {
-               if (!strcmp(bios_settings[i].vendor, vendor) &&
-                               !strcmp(bios_settings[i].version, version)) {
-                       bios_version = i;
+       for (i = 0; bios_settings_table[i].version[0]; ++i) {
+               if (!strcmp(bios_settings_table[i].vendor, vendor) &&
+                   !strcmp(bios_settings_table[i].version, version)) {
+                       bios_settings = &bios_settings_table[i];
                        break;
                }
        }
-       if (bios_version == -1) {
-               pr_err("cannot find BIOS version\n");
+       if (!bios_settings) {
+               pr_err("unknown (unsupported) BIOS version %s/%s, "
+                       "please report, aborting!\n", vendor, version);
                return ACERHDF_ERROR;
        }
        return 0;
Signature · 2 lines
Regards/Gruss,
Boris
author maintainer
#9 lore raw

Re: [PATCH/RFC] Acer Aspire One fan control resume fix, improvements

Hi Andreas, Hi Boris,

I just returned from my holiday trip to italy. Thank you very much for your changes. I'll merge and review them within next 2 days and send a complete patch against linus-git.

How long is the merge window for 2.6.31 opened? Do you think we will finally get the driver merged?

kind regards,
--peter

Quoted 336 lines · Borislav Petkov

Hi,

Quoted 7 lines · Andreas Mohr

Hello all,

tried version 0.5.6, didn't (quite!) like it. ;)
Reason being that when suspending, every couple times there was a fan state check error on resume, leading to the annoying issue of kernel mode fan control getting disabled (via a safety measure).

Quoted 13 lines

ChangeLog:
- reduce maximum temperature check interval, 20 seconds seems a bit much (we have to take into account _external_ heat sources, too!)
- configure BIOS mode during suspend downtime, since our driver code is having a day off (and go back to previous setting after resume)
- update fan state variable during resume, to reflect new state after powering up
- optimization: add bios_settings pointer to current bios's settings, avoids array indirection
- change cmd_off, cmd_auto to fancmd[2] for streamlined code
- several improved log messages
- reverse fan state printing (every value that doesn't indicate "OFF" should definitely lead towards an "AUTO" fan setting!!!!)

good catch!

Quoted 64 lines

 [some functional check was unsafe in this respect, too]
- use LONG_MAX instead of open-coded 0x7fffffffl

Suspend tested multiple times (on 2.6.30-rc8), checkpatch.pl:ed. Version 0.5.7 was internal release only.
This patch is intended for Peter mainly, he should decide what to do with that content - better don't commit it yet.

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +9 -7
@@ -13,6 +13,7 @@
  *               - Carlos Corbacho  cathectic (at) gmail.com
  *  o lkml       - Matthew Garrett
  *               - Borislav Petkov
+ *               - Andreas Mohr
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
@@ -52,7 +53,7 @@
  */
  #undef START_IN_KERNEL_MODE
 
-#define VERSION "0.5.6"
+#define VERSION "0.5.8"
 
  /*
  * According to the Atom N270 datasheet,
@@ -62,8 +63,8 @@
  * assume 89°C is critical temperature.
  */
  #define ACERHDF_TEMP_CRIT 89
-#define ACERHDF_FAN_AUTO 1
  #define ACERHDF_FAN_OFF 0
+#define ACERHDF_FAN_AUTO 1
 
  /*
  * No matter what value the user puts into the fanon variable, turn on the fan
@@ -72,17 +73,18 @@
  #define ACERHDF_MAX_FANON 80
 
  /*
- * Maximal interval between two temperature checks is 20 seconds, as the die
- * can get hot really fast under heavy load
+ * Maximum interval between two temperature checks is 15 seconds, as the die
+ * can get hot really fast under heavy load (plus we shouldn't forget about
+ * possible impact of _external_ aggressive sources such as heaters, sun etc.)
  */
-#define ACERHDF_MAX_INTERVAL 20
+#define ACERHDF_MAX_INTERVAL 15
 
  /*
  * As temperatures can be negative, zero or positive, the value indicating
  * an error must be somewhere beyond valid temperature values.
- * 0x7fffffffl - highest possible positive long value should do the job.
+ * LONG_MAX (highest possible positive long value) should do the job.
  */
-#define ACERHDF_ERROR 0x7fffffffl
+#define ACERHDF_ERROR LONG_MAX
 
 
  #ifdef START_IN_KERNEL_MODE

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 line 102

Quoted 5 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +1 -1
@@ -97,7 +99,7 @@
  static unsigned int verbose;
  static unsigned int fanstate = ACERHDF_FAN_AUTO;
  static int disable_kernelmode;
-static int bios_version = -1;
+static int pre_suspend_kernelmode;

ok, this variable is unnecessary:

it is enough to do

if (kernelmode)

	acerhdf_change_fanstate(ACERHDF_FAN_AUTO);

on suspend and since the machine starts with the BIOS in control of the fan, the acerhdf_set_cur_state() thermal callback will figure out what to do so you don't need to explicitly save the kernelmode setting before suspend and restore it on resume.

Quoted 179 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +40 -32
  static char force_bios[16];
  static unsigned int prev_interval;
  struct thermal_zone_device *acerhdf_thz_dev;
@@ -123,25 +125,26 @@
        const char *version;
        unsigned char fanreg;
        unsigned char tempreg;
-       unsigned char cmd_off;
-       unsigned char cmd_auto;
+       unsigned char fancmd[2]; /* fan off and auto commands */
  };
 
  /* Register addresses and values for different BIOS versions */
-static const struct bios_settings_t bios_settings[] = {
-       {"Acer", "v0.3109", 0x55, 0x58, 0x1f, 0x00},
-       {"Acer", "v0.3114", 0x55, 0x58, 0x1f, 0x00},
-       {"Acer", "v0.3301", 0x55, 0x58, 0xaf, 0x00},
-       {"Acer", "v0.3304", 0x55, 0x58, 0xaf, 0x00},
-       {"Acer", "v0.3305", 0x55, 0x58, 0xaf, 0x00},
-       {"Acer", "v0.3308", 0x55, 0x58, 0x21, 0x00},
-       {"Acer", "v0.3309", 0x55, 0x58, 0x21, 0x00},
-       {"Acer", "v0.3310", 0x55, 0x58, 0x21, 0x00},
-       {"Gateway", "v0.3103", 0x55, 0x58, 0x21, 0x00},
-       {"Packard Bell", "v0.3105", 0x55, 0x58, 0x21, 0x00},
-       {"", 0, 0, 0, 0, 0}
+static const struct bios_settings_t bios_settings_table[] = {
+       {"Acer", "v0.3109", 0x55, 0x58, {0x1f, 0x00} },
+       {"Acer", "v0.3114", 0x55, 0x58, {0x1f, 0x00} },
+       {"Acer", "v0.3301", 0x55, 0x58, {0xaf, 0x00} },
+       {"Acer", "v0.3304", 0x55, 0x58, {0xaf, 0x00} },
+       {"Acer", "v0.3305", 0x55, 0x58, {0xaf, 0x00} },
+       {"Acer", "v0.3308", 0x55, 0x58, {0x21, 0x00} },
+       {"Acer", "v0.3309", 0x55, 0x58, {0x21, 0x00} },
+       {"Acer", "v0.3310", 0x55, 0x58, {0x21, 0x00} },
+       {"Gateway", "v0.3103", 0x55, 0x58, {0x21, 0x00} },
+       {"Packard Bell", "v0.3105", 0x55, 0x58, {0x21, 0x00} },
+       {"", 0, 0, 0, {0, 0} }
  };
 
+static const struct bios_settings_t *bios_settings __read_mostly;
+
 
  /* acer ec functions */
  static int acerhdf_get_temp(void)
@@ -149,7 +152,7 @@
        u8 temp;
 
        /* read temperature */
-       if (!ec_read(bios_settings[bios_version].tempreg, &temp)) {
+       if (!ec_read(bios_settings->tempreg, &temp)) {
                if (verbose)
                        pr_notice("temp %d\n", temp);
                return temp;
@@ -161,8 +164,8 @@
  {
        u8 fan;
 
-       if (!ec_read(bios_settings[bios_version].fanreg, &fan))
-               return (fan == bios_settings[bios_version].cmd_off) ?
+       if (!ec_read(bios_settings->fanreg, &fan))
+               return (fan == bios_settings->fancmd[ACERHDF_FAN_OFF]) ?
                        ACERHDF_FAN_OFF : ACERHDF_FAN_AUTO;
 
        return ACERHDF_ERROR;
@@ -173,19 +176,19 @@
        unsigned char cmd;
 
        if (verbose)
-               pr_notice("fan %s\n", (state == ACERHDF_FAN_AUTO) ?
-                               "ON" : "OFF");
+               pr_notice("fan %s\n", (state == ACERHDF_FAN_OFF) ?
+                               "OFF" : "ON");
 
-       if (state == ACERHDF_FAN_AUTO) {
-               cmd = bios_settings[bios_version].cmd_auto;
-               fanstate = ACERHDF_FAN_AUTO;
-       } else {
-               cmd = bios_settings[bios_version].cmd_off;
-               fanstate = ACERHDF_FAN_OFF;
+       if ((state != ACERHDF_FAN_OFF) && (state != ACERHDF_FAN_AUTO)) {
+               pr_err("invalid fan state %d requested, setting to auto!\n",
+                       state);
+               state = ACERHDF_FAN_AUTO;
        }
 
-       ec_write(bios_settings[bios_version].fanreg, cmd);
+       cmd = bios_settings->fancmd[state];
+       fanstate = state;
 
+       ec_write(bios_settings->fanreg, cmd);
  }
 
  /* helpers */
@@ -253,6 +256,18 @@
        return 0;
  }
 
+/* provide one central function to set disable_kernelmode
+ * (always set ACERHDF_FAN_AUTO, too!) */
+static inline void acerhdf_revert_to_bios_mode(void)
+{
+       acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
+       /*
+        * let the thermal layer disable kernel mode. This ensures that
+        * the thermal layer doesn't switch off the fan again
+        */
+       disable_kernelmode = 1;
+}
+
  /*  current operation mode - enabled / disabled */
  static int acerhdf_get_mode(struct thermal_zone_device *thermal,
                enum thermal_device_mode *mode)
@@ -279,13 +294,8 @@
                pr_notice("kernel mode OFF\n");
                thermal->polling_delay = 0;
                thermal_zone_device_update(thermal);
-               acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
 
-               /*
-                * let the thermal layer disable kernel mode. This ensures that
-                * the thermal layer doesn't switch off the fan again
-                */
-               disable_kernelmode = 1;
+               acerhdf_revert_to_bios_mode();
        } else {
                kernelmode = 1;
                pr_notice("kernel mode ON\n");
@@ -394,21 +404,19 @@
         */
        if (cur_state != fanstate) {

               pr_err("failed reading fan state, "
-                               "disabling kernelmode.\n");
+                               "falling back to default BIOS handling.\n"); pr_err("read state: %d expected state: %d\n", cur_state, fanstate);

-               acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
-               disable_kernelmode = 1;
+               acerhdf_revert_to_bios_mode();
               return -EINVAL;
       }
       /* same with temperature */
       if (cur_temp == ACERHDF_ERROR) {
               pr_err("failed reading temperature, "
-                               "disabling kernelmode.\n");
+                               "falling back to default BIOS handling.\n");

-               acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
-               disable_kernelmode = 1;
+               acerhdf_revert_to_bios_mode(); return -EINVAL;
       }

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +13 -3
@@ -437,11 +445,21 @@
                pm_message_t state)
  {
        /*
-        * in kernelmode turn on fan, because the aspire one awakes with
-        * spinning fan
+        * always revert to BIOS mode during suspend/resume activities:
+        * a) during suspend our driver is inactive, thus if there's
+        *    anything to be done fan-wise, it's the BIOS's job...
+        * b) Aspire awakes with spinning fan in BIOS mode,
+        *    thus we better do the same (behaviour is preserved if we use BIOS)
         */
-       if (kernelmode)
+
+       /* remember previous setting */
+       pre_suspend_kernelmode = kernelmode;
+
+       if (kernelmode) {
                acerhdf_change_fanstate(ACERHDF_FAN_AUTO);
+               kernelmode = 0;
+       }
+
        if (verbose)
                pr_notice("going suspend\n");
        return 0;

drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 line 475

Quoted 9 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +6 -0
@@ -449,6 +467,14 @@
 
  static int acerhdf_resume(struct platform_device *device)
  {
+       kernelmode = pre_suspend_kernelmode;
+
+       /* update our fanstate variable to the possibly different
+        * post-resume fan state
+        * (to prevent a safety check from failing)
+        */

please use kernel-style comments:

/*
* comment text goes here
*/

Quoted 29 lines
drivers/platform/x86/acerhdf.c     2009-06-12 16:10:58.000000000 +0200 renamed from drivers/platform/x86/acerhdf.c.patched_orig        2009-06-12 10:39:30.000000000 +0200 +9 -6
+       fanstate = acerhdf_get_fanstate();
+
        if (verbose)
                pr_notice("resuming\n");
        return 0;
@@ -526,15 +552,16 @@
 
 
        /* search BIOS version and BIOS vendor in BIOS settings table */
-       for (i = 0; bios_settings[i].version[0]; ++i) {
-               if (!strcmp(bios_settings[i].vendor, vendor) &&
-                               !strcmp(bios_settings[i].version, version)) {
-                       bios_version = i;
+       for (i = 0; bios_settings_table[i].version[0]; ++i) {
+               if (!strcmp(bios_settings_table[i].vendor, vendor) &&
+                   !strcmp(bios_settings_table[i].version, version)) {
+                       bios_settings = &bios_settings_table[i];
                        break;
                }
        }
-       if (bios_version == -1) {
-               pr_err("cannot find BIOS version\n");
+       if (!bios_settings) {
+               pr_err("unknown (unsupported) BIOS version %s/%s, "
+                       "please report, aborting!\n", vendor, version);
                return ACERHDF_ERROR;
        }
        return 0;
Signature · 2 lines
Regards/Gruss,
Boris
#10 lore raw

Re: [PATCH/RFC] Acer Aspire One fan control resume fix, improvements

Hi Peter,

Quoted 8 lines · Peter Feuerer

Hi Andreas, Hi Boris,

I just returned from my holiday trip to italy. Thank you very much for your changes. I'll merge and review them within next 2 days and send a complete patch against linus-git.

How long is the merge window for 2.6.31 opened? Do you think we will finally get the driver merged?

the merge window opened sometime around mid last week so I think there's plenty of time. However, it is always a good idea if the driver sees a bit more testing by a wider audience, maybe linux-next.

Andrew, what do you think?

Signature · 2 lines
Regards/Gruss,
    Boris.
#11 lore raw

Re: [PATCH/RFC] Acer Aspire One fan control resume fix, improvements

Quoted 9 lines · On Tuesday 16 June 2009 02:01:07 Borislav Petkov wrote:

Hi Peter,

Quoted 6 lines · Peter Feuerer

Hi Andreas, Hi Boris,

I just returned from my holiday trip to italy. Thank you very much for your changes. I'll merge and review them within next 2 days and send a complete patch against linus-git.

Quoted 8 lines
Quoted 2 lines

How long is the merge window for 2.6.31 opened? Do you think we will finally get the driver merged?

the merge window opened sometime around mid last week so I think there's plenty of time. However, it is always a good idea if the driver sees a bit more testing by a wider audience, maybe linux-next.

Andrew, what do you think?

Please add this to the kernel now. I have been using version of this on my wife's acer since I installed linux on it. Its been doing its job as adverstised. This is not a core feature - its a driver. It works. Why not add it?

Ed Tomlinson

#12 lore raw

Re: [PATCH/RFC] Acer Aspire One fan control resume fix, improvements

Hi,

Quoted 3 lines · Ed Tomlinson

Please add this to the kernel now. I have been using version of this on my wife's acer since I installed linux on it. Its been doing its job as adverstised. This is not a core feature - its a driver. It works. Why not add it?

Seconded. Better not wait 18 months this time.

This being said, I'm unfortunately _still_ experiencing the resume issue (unexpected fan state) from time to time after resume. Really needs more investigation
to get that fixed somehow, eventually.

Also, the pre_suspend_kernelmode variable is superfluous indeed, as mentioned this can be simplified.

Atom emergency shutdown works fine on my machine (blocked fan sometimes, going tick-tick-tick on boot etc., nice way to test this BIOS feature ;), thus given this extra protection I believe the driver can be committed as is anyway, nevermind any remaining driver issues which _might_ perhaps eventually still exist.

Andreas Mohr

author maintainer
#13 lore raw

Hi,

Andreas, I applied and reviewd your patch and additionally changed following things:
o added "acerhdf_enable_kernelmode" which is the opposite of your "acerhdf_revert_to_bios_mode".
o suspend stops and resume starts the kernelmode in a clean way
o some small bugfixes (e.g. Version of the terminator bios table entry has to be "" and not 0)...

I tested the patch with current linus git. Unfortunately does the current state of the kernel freeze my a1 after suspend/resume randomly (also without acerhdf). Anyways I was able to do 11 suspend / resume cycles with acerhdf loaded and kernelmode=on.

Signature · 10 lines
Peter Feuerer <[email protected]>


Acerhdf is a driver for Acer Aspire One netbooks. It allows to access
the temperature sensor and to control the fan.

Signed-off-by: Peter Feuerer <[email protected]>
Signed-off-by: Andreas Mohr <[email protected]>
Reviewed-by: Borislav Petkov <[email protected]>
Tested-by: Borislav Petkov <[email protected]>
MAINTAINERS +7 -0
index 09f6b3e..dc3e4fb 100644
@@ -224,6 +224,13 @@ L:	[email protected]
 S:	Maintained
 F:	drivers/net/acenic*
 
+ACER ASPIRE ONE TEMPERATURE AND FAN DRIVER
+P: Peter Feuerer
+M: [email protected]
+W: http://piie.net/?section=acerhdf
+S: Maintained
+F: drivers/platform/x86/acerhdf.c
+
 ACER WMI LAPTOP EXTRAS
 P:	Carlos Corbacho
 M:	[email protected]
drivers/platform/x86/Kconfig +17 -0
index c682ac5..c768475 100644
@@ -34,6 +34,23 @@ config ACER_WMI
 	  If you have an ACPI-WMI compatible Acer/ Wistron laptop, say Y or M
 	  here.
 
+config ACERHDF
+	tristate "Acer Aspire One temperature and fan driver"
+	depends on THERMAL && THERMAL_HWMON
+	---help---
+	  This is a driver for Acer Aspire One netbooks. It allows to access
+	  the temperature sensor and to control the fan.
+
+	  After loading this driver the BIOS is still in control of the fan.
+	  To let the kernel handle the fan, do:
+	  echo -n enabled > /sys/class/thermal/thermal_zone0/mode
+
+	  For more information about this driver see
+	  <http://piie.net/files/acerhdf_README.txt>
+
+	  If you have an Acer Aspire One netbook, say Y or M
+	  here.
+
 config ASUS_LAPTOP
 	tristate "Asus Laptop Extras (EXPERIMENTAL)"
 	depends on ACPI
drivers/platform/x86/Makefile +1 -0
index e40c7bd..641b8bf 100644
@@ -9,6 +9,7 @@ obj-$(CONFIG_COMPAL_LAPTOP)	+= compal-laptop.o
 obj-$(CONFIG_DELL_LAPTOP)	+= dell-laptop.o
 obj-$(CONFIG_DELL_WMI)		+= dell-wmi.o
 obj-$(CONFIG_ACER_WMI)		+= acer-wmi.o
+obj-$(CONFIG_ACERHDF)		+= acerhdf.o
 obj-$(CONFIG_HP_WMI)		+= hp-wmi.o
 obj-$(CONFIG_TC1100_WMI)	+= tc1100-wmi.o
 obj-$(CONFIG_SONY_LAPTOP)	+= sony-laptop.o
drivers/platform/x86/acerhdf.c new file +663 -0

663 lines not on this page · raw message

#14 lore raw
Quoted 27 lines · On Wed, 17 Jun 2009 00:14:35 +0200 Peter Feuerer wrote:

Hi,

Andreas, I applied and reviewd your patch and additionally changed following things:
o added "acerhdf_enable_kernelmode" which is the opposite of your "acerhdf_revert_to_bios_mode".
o suspend stops and resume starts the kernelmode in a clean way
o some small bugfixes (e.g. Version of the terminator bios table entry has to be "" and not 0)...

I tested the patch with current linus git. Unfortunately does the current state of the kernel freeze my a1 after suspend/resume randomly (also without acerhdf). Anyways I was able to do 11 suspend / resume cycles with acerhdf loaded and kernelmode=on.

Signature · 10 lines
Peter Feuerer <[email protected]>


Acerhdf is a driver for Acer Aspire One netbooks. It allows to access
the temperature sensor and to control the fan.

Signed-off-by: Peter Feuerer <[email protected]>
Signed-off-by: Andreas Mohr <[email protected]>
Reviewed-by: Borislav Petkov <[email protected]>
Tested-by: Borislav Petkov <[email protected]>
drivers/platform/x86/Kconfig +0 -0
index c682ac5..c768475 100644

drivers/platform/x86/Kconfig config ACER_WMI · line 39

Quoted 6 lines
drivers/platform/x86/Kconfig +3 -0
@@ -34,6 +34,23 @@ config ACER_WMI
 	  If you have an ACPI-WMI compatible Acer/ Wistron laptop, say Y or M
 	  here.
 
+config ACERHDF
+	tristate "Acer Aspire One temperature and fan driver"
+	depends on THERMAL && THERMAL_HWMON

depends on ACPI

due to calling ec_read()

Quoted 7 lines
drivers/platform/x86/Kconfig +7 -0
+	---help---
+	  This is a driver for Acer Aspire One netbooks. It allows to access
+	  the temperature sensor and to control the fan.
+
+	  After loading this driver the BIOS is still in control of the fan.
+	  To let the kernel handle the fan, do:
+	  echo -n enabled > /sys/class/thermal/thermal_zone0/mode

drivers/platform/x86/Kconfig config ACER_WMI · line 56

Quoted 10 lines
drivers/platform/x86/Kconfig +7 -0
+
+	  For more information about this driver see
+	  <http://piie.net/files/acerhdf_README.txt>
+
+	  If you have an Acer Aspire One netbook, say Y or M
+	  here.
+
 config ASUS_LAPTOP
 	tristate "Asus Laptop Extras (EXPERIMENTAL)"
 	depends on ACPI

---
~Randy
LPC 2009, Sept. 23-25, Portland, Oregon
http://linuxplumbersconf.org/2009/

#15 lore raw
In reply to Peter Feuerer

Hi,

Quoted 4 lines · Peter Feuerer

Hi,

Andreas, I applied and reviewd your patch and additionally changed following things:
o added "acerhdf_enable_kernelmode" which is the opposite of your "acerhdf_revert_to_bios_mode".

Good idea.

Quoted 1 line

o suspend stops and resume starts the kernelmode in a clean way

Didn't someone argue that pre_suspend_kernelmode was unnecessary? It's still in there. (could be submitted together with the Kconfig ACPI depends ;)

Quoted 1 line

I tested the patch with current linus git. Unfortunately does the current state of the kernel freeze my a1 after suspend/resume randomly (also without acerhdf). Anyways I was able to do 11 suspend / resume cycles with acerhdf loaded and kernelmode=on.

Oh no, not again.

Thanks!

Andreas Mohr

author maintainer
#16 lore raw

Hi Andreas, Hi Boris,

Quoted 5 lines · Andreas Mohr
Quoted 1 line

o suspend stops and resume starts the kernelmode in a clean way

Didn't someone argue that pre_suspend_kernelmode was unnecessary? It's still in there. (could be submitted together with the Kconfig ACPI depends ;)

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

I'll add the ACPI dependency to Kconfig and submit the patch to lkml and directly to Linus, as Len doesn't seem to react?! - Boris what do you think?

I mean, even if there's a problem with suspend / resume, this can also be fixed after the merge window, while all those release candidates are done.

cheers,
--peter

#17 lore raw

Hi,

Quoted 11 lines · Peter Feuerer

Hi Andreas, Hi Boris,

Quoted 5 lines · Andreas Mohr
Quoted 1 line

o suspend stops and resume starts the kernelmode in a clean way

Didn't someone argue that pre_suspend_kernelmode was unnecessary? It's still in there. (could be submitted together with the Kconfig ACPI depends ;)

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

and it is needed, because...?

Quoted 5 lines

I'll add the ACPI dependency to Kconfig and submit the patch to lkml and directly to Linus, as Len doesn't seem to react?! - Boris what do you think? I mean, even if there's a problem with suspend / resume, this can also be fixed after the merge window, while all those release candidates are done.

Nope, this has to be fixed before submission. The merge window is open, I think, until mid next week so there's plenty of time.

Signature · 2 lines
Regards/Gruss,
Boris
author maintainer
#18 lore raw
Quoted 9 lines · Borislav Petkov

Hi,

Quoted 6 lines · Peter Feuerer

Hi Andreas, Hi Boris,

Quoted 2 lines · Andreas Mohr
Quoted 1 line

o suspend stops and resume starts the kernelmode in a clean way

Quoted 7 lines
Quoted 5 lines
Quoted 3 lines

Didn't someone argue that pre_suspend_kernelmode was unnecessary? It's still in there. (could be submitted together with the Kconfig ACPI depends ;)

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

and it is needed, because...?

It's needed because we do now a clean revert to bios mode before we suspend. And after resume we have to switch to kernelmode again, if the driver was in kernelmode before suspend. So we need to keep track of in what state the driver was before suspending. That's what's this variable is for.

Quoted 8 lines
Quoted 5 lines

I'll add the ACPI dependency to Kconfig and submit the patch to lkml and directly to Linus, as Len doesn't seem to react?! - Boris what do you think? I mean, even if there's a problem with suspend / resume, this can also be fixed after the merge window, while all those release candidates are done.

Nope, this has to be fixed before submission. The merge window is open, I think, until mid next week so there's plenty of time.

I'll be only there until Sunday evening. I'm going to be on business trip starting on Monday (and returning probably at Saturday). But I'll do my best to figure out the problem.

I wasn't able to reproduce the suspend / resume problem Andreas has with acerhdf 0.5.9 yesterday and the day before. Do you also have the problem, that acerhdf is reading sometimes an unexpected Fan state after resuming?

kind regards,
--peter

#19 lore raw

Hi,

Quoted 7 lines · Peter Feuerer
Quoted 7 lines
Quoted 6 lines
Quoted 3 lines

Didn't someone argue that pre_suspend_kernelmode was unnecessary? It's still in there. (could be submitted together with the Kconfig ACPI depends ;)

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

Quoted 6 lines
Quoted 1 line

and it is needed, because...?

It's needed because we do now a clean revert to bios mode before we suspend. And after resume we have to switch to kernelmode again, if the driver was in kernelmode before suspend. So we need to keep track of in what state the driver was before suspending. That's what's this variable is for.

You've got that state in the 'kernelmode' variable. See full comment:
http://marc.info/?l=linux-kernel&m=124482114200865

Quoted 9 lines
Quoted 8 lines
Quoted 5 lines

I'll add the ACPI dependency to Kconfig and submit the patch to lkml and directly to Linus, as Len doesn't seem to react?! - Boris what do you think? I mean, even if there's a problem with suspend / resume, this can also be fixed after the merge window, while all those release candidates are done.

Nope, this has to be fixed before submission. The merge window is open, I think, until mid next week so there's plenty of time.

Quoted 7 lines

I'll be only there until Sunday evening. I'm going to be on business trip starting on Monday (and returning probably at Saturday). But I'll do my best to figure out the problem.

I wasn't able to reproduce the suspend / resume problem Andreas has with acerhdf 0.5.9 yesterday and the day before. Do you also have the problem, that acerhdf is reading sometimes an unexpected Fan state after resuming?

Will look into it on the weekend.

Signature · 2 lines
Regards/Gruss,
Boris
author maintainer
#20 lore raw

Hi,

Quoted 12 lines · Borislav Petkov

Hi,

Quoted 9 lines · Peter Feuerer
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Didn't someone argue that pre_suspend_kernelmode was unnecessary? It's still in there. (could be submitted together with the Kconfig ACPI depends ;)

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

and it is needed, because...?

Quoted 7 lines
Quoted 4 lines

It's needed because we do now a clean revert to bios mode before we suspend. And after resume we have to switch to kernelmode again, if the driver was in kernelmode before suspend. So we need to keep track of in what state the driver was before suspending. That's what's this variable is for.

You've got that state in the 'kernelmode' variable. See full comment:
http://marc.info/?l=linux-kernel&m=124482114200865

We are talking about patch 0.5.9 and not 0.5.8, are we?
http://patchwork.kernel.org/patch/30733/mbox/

have a look at at line 543:
+ /* remember previous setting */
+ pre_suspend_kernelmode = kernelmode;
+
+ if (kernelmode) {
+ acerhdf_revert_to_bios_mode();
+ if (acerhdf_thz_dev)
+ thermal_zone_device_update(acerhdf_thz_dev);
+ }

o we save the kernelmode to pre_suspend_kernelmode
o then if kernelmode is on, we call acerhdf_revert_to_bios_mode()
o we start a thermal_zone_device_update which let the thermal layer call acerhdf_set_cur_state
o acerhdf_set_cur_state then disables kernelmode (and sets kernelmode = 0) so we would lose the state of the kernelmode if we didn't save it within pre_suspend_kernelmode

Quoted 14 lines
Quoted 14 lines
Quoted 9 lines
Quoted 6 lines

I'll add the ACPI dependency to Kconfig and submit the patch to lkml and directly to Linus, as Len doesn't seem to react?! - Boris what do you think? I mean, even if there's a problem with suspend / resume, this can also be fixed after the merge window, while all those release candidates are done.

Nope, this has to be fixed before submission. The merge window is open, I think, until mid next week so there's plenty of time.

I'll be only there until Sunday evening. I'm going to be on business trip starting on Monday (and returning probably at Saturday). But I'll do my best to figure out the problem.

Quoted 5 lines
Quoted 3 lines

I wasn't able to reproduce the suspend / resume problem Andreas has with acerhdf 0.5.9 yesterday and the day before. Do you also have the problem, that acerhdf is reading sometimes an unexpected Fan state after resuming?

Will look into it on the weekend.

Great! Thank you very much!

regards,
--peter

#21 lore raw

Hi,

Quoted 18 lines · Peter Feuerer
Quoted 12 lines
Quoted 11 lines
Quoted 4 lines
Quoted 2 lines

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

and it is needed, because...?

It's needed because we do now a clean revert to bios mode before we suspend.
And after resume we have to switch to kernelmode again, if the driver was in
kernelmode before suspend. So we need to keep track of in what state the driver was before suspending. That's what's this variable is for.

Quoted 2 lines

You've got that state in the 'kernelmode' variable. See full comment:
http://marc.info/?l=linux-kernel&m=124482114200865

We are talking about patch 0.5.9 and not 0.5.8, are we?
http://patchwork.kernel.org/patch/30733/mbox/

Quoted 9 lines

have a look at at line 543:
+       /* remember previous setting */
+       pre_suspend_kernelmode = kernelmode;
+
+       if (kernelmode) {
+               acerhdf_revert_to_bios_mode();
+               if (acerhdf_thz_dev)
+                       thermal_zone_device_update(acerhdf_thz_dev);
+       }

ok, this starts to look quite a bit overengineered for no reason. First, acerhdf_revert_to_bios_mode() sets the fan to auto. Then, you've added a thermal_zone_device_update() call in there which does set the fan to auto indirectly _again_. And we end up with _three_ variables which represent only _one_ state. Here's what it should do:

suspend:
- set fan to auto

resume:
- the thermal layer figures out what to do based on the 'kernelmode' and current temperature.

That's it, everything else is too much.

I'll have a deeper look during the weekend.

Signature · 2 lines
Regards/Gruss,
Boris
#22 lore raw

Hi,

Quoted 6 lines · Borislav Petkov

ok, this starts to look quite a bit overengineered for no reason. First, acerhdf_revert_to_bios_mode() sets the fan to auto. Then, you've added a thermal_zone_device_update() call in there which does set the fan to auto indirectly _again_. And we end up with _three_ variables which represent only _one_ state. Here's what it should do:

Quoted 8 lines

suspend:
- set fan to auto

resume:
- the thermal layer figures out what to do based on the 'kernelmode' and current temperature.

That's it, everything else is too much.

No, I'd say it's not as easy as that.
If you want to discover any weird state changes behind your back (and we want that, that driver should be _safe_), then we need to record the current fan state as set by us.

But that then means that during interruption of normal operation (suspend), we need to temporarily stop this checking.

IOW, we need some more variable(s) to indicate:
OPERATION_SUSPENDED (set by .suspend)
OPERATION_ABOUT_TO_CONTINUE (set by .resume)
OPERATING

In state OPERATION_ABOUT_TO_CONTINUE the thermal layer function would know that they shouldn't check fan state consistency, would refresh fan state to the current state and then set it to OPERATING.

Or some such (would need further analysis).

Andreas Mohr

author maintainer Reviewed-by Borislav Petkov Tested-by Borislav Petkov
#23 lore raw

Hi,

new patch attached, changes:
o probably fixed suspend/resume problem (tested about 100 times with official 2.6.30 release)
o changed temperature notice in "acerhdf_get_temp" to allow suppression of the pr_notice. In verbose mode was the temperature printed out two times, as the thermal layer first called "acerhdf_get_ec_temp" and then called "acerhdf_set_cur_state" which both read the temperature. The temperature read in "acerhdf_set_cur_state" omits now pr_notice of "acerhdf_get_temp".
o added ACPI dependency to Kconfig

As always tested with latest Linus git. (latest Linus git has a problem with suspend/resume, that's why I tested suspend/resume with official 2.6.30)

Boris, Andreas, may you test if suspend/resume works now reliable?

Thank you very much!

kind regards,
peter

---
Acerhdf is a driver for Acer Aspire One netbooks. It allows to access the temperature sensor and to control the fan.

MAINTAINERS +7 -0
index fb94add..0d60560 100644
@@ -230,6 +230,13 @@ L:	[email protected]
 S:	Maintained
 F:	drivers/net/acenic*
 
+ACER ASPIRE ONE TEMPERATURE AND FAN DRIVER
+P: Peter Feuerer
+M: [email protected]
+W: http://piie.net/?section=acerhdf
+S: Maintained
+F: drivers/platform/x86/acerhdf.c
+
 ACER WMI LAPTOP EXTRAS
 P:	Carlos Corbacho
 M:	[email protected]
drivers/platform/x86/Kconfig +17 -0
index c682ac5..5613483 100644
@@ -34,6 +34,23 @@ config ACER_WMI
 	  If you have an ACPI-WMI compatible Acer/ Wistron laptop, say Y or M
 	  here.
 
+config ACERHDF
+	tristate "Acer Aspire One temperature and fan driver"
+	depends on THERMAL && THERMAL_HWMON && ACPI
+	---help---
+	  This is a driver for Acer Aspire One netbooks. It allows to access
+	  the temperature sensor and to control the fan.
+
+	  After loading this driver the BIOS is still in control of the fan.
+	  To let the kernel handle the fan, do:
+	  echo -n enabled > /sys/class/thermal/thermal_zone0/mode
+
+	  For more information about this driver see
+	  <http://piie.net/files/acerhdf_README.txt>
+
+	  If you have an Acer Aspire One netbook, say Y or M
+	  here.
+
 config ASUS_LAPTOP
 	tristate "Asus Laptop Extras (EXPERIMENTAL)"
 	depends on ACPI
drivers/platform/x86/Makefile +1 -0
index e40c7bd..641b8bf 100644
@@ -9,6 +9,7 @@ obj-$(CONFIG_COMPAL_LAPTOP)	+= compal-laptop.o
 obj-$(CONFIG_DELL_LAPTOP)	+= dell-laptop.o
 obj-$(CONFIG_DELL_WMI)		+= dell-wmi.o
 obj-$(CONFIG_ACER_WMI)		+= acer-wmi.o
+obj-$(CONFIG_ACERHDF)		+= acerhdf.o
 obj-$(CONFIG_HP_WMI)		+= hp-wmi.o
 obj-$(CONFIG_TC1100_WMI)	+= tc1100-wmi.o
 obj-$(CONFIG_SONY_LAPTOP)	+= sony-laptop.o
drivers/platform/x86/acerhdf.c new file +663 -0

663 lines not on this page · raw message

#24 lore raw
In reply to Borislav Petkov

Hi,

Quoted 31 lines · Borislav Petkov

Hi,

Quoted 18 lines · Peter Feuerer
Quoted 12 lines
Quoted 11 lines
Quoted 4 lines
Quoted 2 lines

Actually I think pre_suspend_kernelmode is needed, so it won't be dropped.

and it is needed, because...?

It's needed because we do now a clean revert to bios mode before we suspend.
And after resume we have to switch to kernelmode again, if the driver was in
kernelmode before suspend. So we need to keep track of in what state the driver was before suspending. That's what's this variable is for.

Quoted 2 lines

You've got that state in the 'kernelmode' variable. See full comment:
http://marc.info/?l=linux-kernel&m=124482114200865

We are talking about patch 0.5.9 and not 0.5.8, are we?
http://patchwork.kernel.org/patch/30733/mbox/

Quoted 9 lines

have a look at at line 543:
+       /* remember previous setting */
+       pre_suspend_kernelmode = kernelmode;
+
+       if (kernelmode) {
+               acerhdf_revert_to_bios_mode();
+               if (acerhdf_thz_dev)
+                       thermal_zone_device_update(acerhdf_thz_dev);
+       }

Quoted 5 lines

ok, this starts to look quite a bit overengineered for no reason. First, acerhdf_revert_to_bios_mode() sets the fan to auto. Then, you've added a thermal_zone_device_update() call in there which does set the fan to auto indirectly _again_. And we end up with _three_ variables which represent only _one_ state. Here's what it should do:

You are partly right, setting the fan to auto in "acerhdf_revert_to_bios_mode" can be removed, as this is done by the thermal layer when calling "acerhdf_set_cur_state" with disable_kernelmode=1.
But besides that I think it is a clean way. This way we _completely_ disable the thermal polling before going suspend and ensure the thermal layer doesn't handle the fan anymore. When we resume we let the thermal layer take over the fan again. In my opinion this is much cleaner than just switching the fan to auto without keeping the polling of the thermal layer in mind. The big problem with the polling is, that you don't know when the next thermal polling shot arrives. Is it after acerhdf_suspend was called, or after system-resume but before acerhdf_resume was called, or after acerhdf_resume was called… You can't know! That's why in my opinion completely disabling our kernelmode is the only clean solution.

--peter

P.S. I built the official 2.6.30 release (because current git is freezing all the time), patched it with acerhdf 0.5.9 and was testing suspend/resume about 40 times now. Was always hitting the suspend button while working ;). I wasn't able to reproduce the "unexpected fanstate" issue, <=0.5.8 had.

#25 lore raw

Hi,

Quoted 4 lines · Peter Feuerer

You are partly right, setting the fan to auto in "acerhdf_revert_to_bios_mode" can be removed, as this is done by the thermal layer when calling "acerhdf_set_cur_state" with disable_kernelmode=1.

No, it was done _specifically_ this way to make sure that _exactly when_ switching away from kernel mode FAN_AUTO _always_ (by asking people to always call this central function) gets set, too.
I don't want to depend on a separate, _uncontrollable_ entity (thermal layer function) to have to run sometime later
in order to set the fan to FAN_AUTO (IOW, BIOS-controlled mode) accordingly. _Of course_ this FAN_AUTO call is redundant in _most_ cases, but very intentionally so. Anything else would be woefully unsafe (with a fan remaining in FAN_OFF position until machine meltdown).

Andreas Mohr

#26 lore raw

Hi,

Quoted 5 lines · Andreas Mohr

Hi,

Quoted 1 line · Peter Feuerer

You are partly right, setting the fan to auto in "acerhdf_revert_to_bios_mode" can be removed, as this is done by the thermal layer when calling "acerhdf_set_cur_state" with disable_kernelmode=1.

Quoted 9 lines

No, it was done _specifically_ this way to make sure that _exactly when_ switching away from kernel mode FAN_AUTO _always_ (by asking people to always call this central function) gets set, too.
I don't want to depend on a separate, _uncontrollable_ entity (thermal layer function) to have to run sometime later
in order to set the fan to FAN_AUTO (IOW, BIOS-controlled mode) accordingly. _Of course_ this FAN_AUTO call is redundant in _most_ cases, but very intentionally so. Anything else would be woefully unsafe (with a fan remaining in FAN_OFF position until machine meltdown).

Ok, that sounds reasonable.

--peter