[lm-sensors] [PATCH] On n2100 systems,

[lm-sensors] [PATCH] On n2100 systems,

[Riku Voipio]

On thecus n2100, the bootloader does not setup fans to run. In order
to protect the user from frying their gear, start up fans on boot.

Depends on:

f75375s: Add new style bindings
f75375s: allow setting up fans with platform_data

patches sent to lm-sensors mailing list.

Signed-off-by: Riku Voipio <riku.voipio@movial.fi>
---
 arch/arm/mach-iop32x/n2100.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/arch/arm/mach-iop32x/n2100.c b/arch/arm/mach-iop32x/n2100.c
index 1873bd8..5a4acff 100644
--- a/arch/arm/mach-iop32x/n2100.c
+++ b/arch/arm/mach-iop32x/n2100.c
@@ -16,6 +16,7 @@
 
 #include <linux/mm.h>
 #include <linux/init.h>
+#include <linux/f75375s.h>
 #include <linux/delay.h>
 #include <linux/kernel.h>
 #include <linux/pci.h>
@@ -200,11 +201,21 @@ static struct platform_device n2100_serial_device = {
 	.resource	= &n2100_uart_resource,
 };
 
+static struct f75375s_platform_data n2100_f75375s = {
+	.pwm		= { 255, 255 },
+	.pwm_enable = { 0, 0 },
+};
+
 static struct i2c_board_info __initdata n2100_i2c_devices[] = {
 	{
 		I2C_BOARD_INFO("rtc-rs5c372", 0x32),
 		.type = "rs5c372b",
 	},
+	{
+		I2C_BOARD_INFO("f75375", 0x2e),
+		.type = "f75375",
+		.platform_data = &n2100_f75375s,
+	},
 };
 
 /*
-- 
1.5.3.1


-- 
"rm -rf" only sounds scary if you don't have backups

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Lennert Buytenhek]

On Tue, Oct 16, 2007 at 01:08:57PM +0300, Riku Voipio wrote:

> On thecus n2100, the bootloader does not setup fans to run. In order
> to protect the user from frying their gear, start up fans on boot.
> 
> Depends on:
> 
> f75375s: Add new style bindings
> f75375s: allow setting up fans with platform_data
> 
> patches sent to lm-sensors mailing list.
> 
> Signed-off-by: Riku Voipio <riku.voipio@movial.fi>

Acked-by: Lennert Buytenhek <buytenh@wantstofly.org>

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Jean Delvare]

Hi Riku,

On Tue, 16 Oct 2007 13:08:57 +0300, Riku Voipio wrote:
> On thecus n2100, the bootloader does not setup fans to run. In order
> to protect the user from frying their gear, start up fans on boot.

Why don't you fix the "bootloader" (whatever it is in your context)
instead? Fixing the fan speed at the OS level sounds wrong. The user
might be using a different OS. The user might even not run any OS.
Would the hardware fry if the user stays in the bootloader?

On top of this, if I read the the datasheet of the F75375S correctly,
it has the fans on (in "speed mode") by default. This suggests that the
"bootloader" is no only "not setting up fans to run" but it is actively
turning them off. So that's what you really need to fix.

> 
> Depends on:
> 
> f75375s: Add new style bindings
> f75375s: allow setting up fans with platform_data
> 
> patches sent to lm-sensors mailing list.
> 
> Signed-off-by: Riku Voipio <riku.voipio@movial.fi>
> ---
>  arch/arm/mach-iop32x/n2100.c |   11 +++++++++++
>  1 files changed, 11 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/arm/mach-iop32x/n2100.c b/arch/arm/mach-iop32x/n2100.c
> index 1873bd8..5a4acff 100644
> --- a/arch/arm/mach-iop32x/n2100.c
> +++ b/arch/arm/mach-iop32x/n2100.c
> @@ -16,6 +16,7 @@
>  
>  #include <linux/mm.h>
>  #include <linux/init.h>
> +#include <linux/f75375s.h>
>  #include <linux/delay.h>
>  #include <linux/kernel.h>
>  #include <linux/pci.h>
> @@ -200,11 +201,21 @@ static struct platform_device n2100_serial_device = {
>  	.resource	= &n2100_uart_resource,
>  };
>  
> +static struct f75375s_platform_data n2100_f75375s = {
> +	.pwm		= { 255, 255 },
> +	.pwm_enable = { 0, 0 },

Bad alignment. And I don't think that PWM can be completely disabled on
the F75375S anyway? So you should set pwm_enable to 1 not 0.

> +};
> +
>  static struct i2c_board_info __initdata n2100_i2c_devices[] = {
>  	{
>  		I2C_BOARD_INFO("rtc-rs5c372", 0x32),
>  		.type = "rs5c372b",
>  	},
> +	{
> +		I2C_BOARD_INFO("f75375", 0x2e),
> +		.type = "f75375",
> +		.platform_data = &n2100_f75375s,
> +	},
>  };
>  
>  /*


-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Lennert Buytenhek]

On Wed, Oct 17, 2007 at 04:57:00PM +0200, Jean Delvare wrote:

> > On thecus n2100, the bootloader does not setup fans to run. In order
> > to protect the user from frying their gear, start up fans on boot.
> 
> Why don't you fix the "bootloader" (whatever it is in your context)

The bootloader is Redboot.  It is shipped as the default bootloader
in the flash of the device as sold, and it runs directly after power
on.  (In that regard, it's kind of like the PC BIOS in x86 machines.)

While I would agree that fixing this in the bootloader is the cleanest
approach, I'm pretty sure that even if we do manage to build a new
version of the bootloader (I'm not even sure whether we have the source
for it), I'm sure many users would be very reluctant to flash a new
bootloader into their device.

We do work around PC BIOS bugs in the Linux kernel as well.  This
isn't a very different situation from that, I think.

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Jean Delvare]

Hi Lennert,

On Wed, 17 Oct 2007 17:19:35 +0200, Lennert Buytenhek wrote:
> On Wed, Oct 17, 2007 at 04:57:00PM +0200, Jean Delvare wrote:
> 
> > > On thecus n2100, the bootloader does not setup fans to run. In order
> > > to protect the user from frying their gear, start up fans on boot.
> > 
> > Why don't you fix the "bootloader" (whatever it is in your context)
> 
> The bootloader is Redboot.  It is shipped as the default bootloader
> in the flash of the device as sold, and it runs directly after power
> on.  (In that regard, it's kind of like the PC BIOS in x86 machines.)

OK, thanks for the clarification.

> While I would agree that fixing this in the bootloader is the cleanest
> approach, I'm pretty sure that even if we do manage to build a new
> version of the bootloader (I'm not even sure whether we have the source
> for it),

Can't you contact whoever sells this system? If the "BIOS" is broken,
that's really up to them to fix it.

>          I'm sure many users would be very reluctant to flash a new
> bootloader into their device.

If their machine fries when they don't, I'm pretty sure they will ;)

> We do work around PC BIOS bugs in the Linux kernel as well.  This
> isn't a very different situation from that, I think.

I'm not aware of such workarounds to prevent machines from frying. For
such critical cases, system vendors typically bother releasing a BIOS
update.

What I would like to understand is how we came in to the current
situation. As I said, stopped fans is not the default for the F75375S
chip. This really suggests that the fans are stopped on purpose by
whoever wrote that Redboot thing. I would appreciate if somebody could
confirm this. Also, are the fans really stopped, of set into automatic
mode and just happen not to spin because the temperature isn't high
enough? Can I see a dump of the chip before the Linux driver is loaded?

So, someone is selling hardware that fries when used because they
disabled the fans on purpose. How can this be? Is this hardware
supposed to be used with another OS, and that other OS does kick the
fans in as you suggest Linux should do? Please clarify the situation.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Lennert Buytenhek]

On Wed, Oct 17, 2007 at 11:42:31PM +0200, Jean Delvare wrote:

> Hi Lennert,

Hey Jean,


> > While I would agree that fixing this in the bootloader is the cleanest
> > approach, I'm pretty sure that even if we do manage to build a new
> > version of the bootloader (I'm not even sure whether we have the source
> > for it),
> 
> Can't you contact whoever sells this system? If the "BIOS" is broken,
> that's really up to them to fix it.

These systems are sold by Thecus.  (www.thecus.com)  The n2100 looks
something like this:

	http://www.trustedreviews.com/images/article/inline/2559-frontleft.jpg

In the default firmware, the bootloader automatically boots an ARM
Linux kernel which contains a fan controller driver.  Meaning, the
default kernel that runs on the machine (which is the only kernel that
the manufacturer has envisioned ever running on the machine -- and most
users of the device won't actually realise that there is a Linux kernel
running on the device at all) handles thermal management correctly.

The n2100 is sold as a purpose-built NAS appliance, not as a
general-purpose ARM hack box.  Even though it's not too hard to hack
the firmware and run your own software on the machine, it's not
something that the product is advertised for.

I'm not sure whether the bootloader explicitly disables the fans or
whether they just don't kick in because the temperature isn't high
enough.


> >          I'm sure many users would be very reluctant to flash a new
> > bootloader into their device.
> 
> If their machine fries when they don't, I'm pretty sure they will ;)

Luckily, their machine won't fry if they don't. :)


> > We do work around PC BIOS bugs in the Linux kernel as well.  This
> > isn't a very different situation from that, I think.
> 
> I'm not aware of such workarounds to prevent machines from frying. For
> such critical cases, system vendors typically bother releasing a BIOS
> update.
> 
> What I would like to understand is how we came in to the current
> situation. As I said, stopped fans is not the default for the F75375S
> chip. This really suggests that the fans are stopped on purpose by
> whoever wrote that Redboot thing.

As I understand, Redboot was written by Red Hat.  It's probably
one of the most widely used ARM bootloaders out there.

The port of Redboot to the IOP 80219 (which is the CPU in the n2100)
was done by Intel, and the port to the n2100 board specifically was
done by Thecus.

As I said above, I am not entirely sure what Redboot does.  Riku, do
you have the source code for the Redboot used in the n2100?


> I would appreciate if somebody could
> confirm this. Also, are the fans really stopped, of set into automatic
> mode and just happen not to spin because the temperature isn't high
> enough?

I'm not entirely sure.


> Can I see a dump of the chip before the Linux driver is loaded?

Riku?


> So, someone is selling hardware that fries when used because they
> disabled the fans on purpose.

It's not as dramatic as that. :)


> How can this be? Is this hardware
> supposed to be used with another OS, and that other OS does kick the
> fans in as you suggest Linux should do? Please clarify the situation.

I don't think other OSes run on this machine.

Well, maybe one of the BSDs do -- but I'm not sure how they handle
the fans.

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Riku Voipio]

Lennert Buytenhek wrote:
>>> While I would agree that fixing this in the bootloader is the cleanest
>>> approach, I'm pretty sure that even if we do manage to build a new
>>> version of the bootloader (I'm not even sure whether we have the source
>>> for it),
>>>       
>> Can't you contact whoever sells this system? If the "BIOS" is broken,
>> that's really up to them to fix it.
>>     
>
>
> I'm not sure whether the bootloader explicitly disables the fans or
> whether they just don't kick in because the temperature isn't high
> enough.
>
>   
Redboot does not touch the fans. The problem is that the fanspeed
reading isn't reliable. The "speed" mode tries to adjust pwm to get
the fanspeed to match whatever is requested in "expect" registers,
but never to enough to get the fans to run.

..
>> I'm not aware of such workarounds to prevent machines from frying. For
>> such critical cases, system vendors typically bother releasing a BIOS
>> update.
>>     
As Lennert mentioned, Software provided by Thecus
does set the fans running by default (by combination of
a kernel driver and userland scripts). It's not sold as a
general purpose PC.

In my experience the risk of literal "frying" is very very low,
but even with just one one disk inside and finnish cool weather,
smartmontools have reported the hard drive going over the
operating temperature limit (60c).

With, say, 2x10 rpm disks inside the case, I'd say the risk of
disks breaking down prematurely is very real. And I do not
want to take the responsibility of people reporting that "installing
Debian caused my hardware to break down"


> As I said above, I am not entirely sure what Redboot does.  Riku, do
> you have the source code for the Redboot used in the n2100?
>   
We have some redboot code from Thecus. But I'm so not going
to touch it, since currently we have no  way to recover the
bootloader in case it goes bust.

Also, "fixing" in this case is more likely to mean "add support
for the fintek chip" rather than "rather"

>> Can I see a dump of the chip before the Linux driver is loaded?
>>     
>
> Riku?
>
>   
I'll grab it, but see the first paragraph I wrote.

>> How can this be? Is this hardware
>> supposed to be used with another OS, and that other OS does kick the
>> fans in as you suggest Linux should do? Please clarify the situation.
>>     
>
> I don't think other OSes run on this machine.
>
> Well, maybe one of the BSDs do -- but I'm not sure how they handle
> the fans.
>   
The one being OpenBSD. They send fans to full speed in
fintek driver with driver paramater. I guess the parameter
gets compiled in the kernel when compiling a n2100 kernel.




_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Jean Delvare]

Hi Riku, Lennert,

On Thu, 18 Oct 2007 11:56:04 +0300, Riku Voipio wrote:
> Redboot does not touch the fans. The problem is that the fanspeed
> reading isn't reliable. The "speed" mode tries to adjust pwm to get
> the fanspeed to match whatever is requested in "expect" registers,
> but never to enough to get the fans to run.

I see. I have to admit that it isn't very smart from Fintek to set the
"speed mode" by default on this chip. It would be safer to default to
full speed. OTOH the default speed target seems to be 4297 RPM if I
read the datasheet correctly, which is high enough. If the fan speed
reading isn't reliable (which admittedly is a bad thing to start with)
I'd expect the chip to set the fan output to the max. Isn't it the case?

> As Lennert mentioned, Software provided by Thecus
> does set the fans running by default (by combination of
> a kernel driver and userland scripts). It's not sold as a
> general purpose PC.
> 
> In my experience the risk of literal "frying" is very very low,
> but even with just one one disk inside and finnish cool weather,
> smartmontools have reported the hard drive going over the
> operating temperature limit (60c).
> 
> With, say, 2x10 rpm disks inside the case, I'd say the risk of
> disks breaking down prematurely is very real. And I do not
> want to take the responsibility of people reporting that "installing
> Debian caused my hardware to break down"

I see. But I can't think of a reason why this can't be simply solved in
user-space, as Thecus is doing in their own product. This would be much
easier than the kernel patches you've been sending.

> > Can I see a dump of the chip before the Linux driver is loaded?
>
> I'll grab it, but see the first paragraph I wrote.

I'm still interested.

Thanks,
-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Riku Voipio]

Jean Delvare wrote:
> Hi Riku, Lennert,
>
> On Thu, 18 Oct 2007 11:56:04 +0300, Riku Voipio wrote:
>> Redboot does not touch the fans. The problem is that the fanspeed
>> reading isn't reliable. The "speed" mode tries to adjust pwm to get
>> the fanspeed to match whatever is requested in "expect" registers,
>> but never to enough to get the fans to run.
>
> I see. I have to admit that it isn't very smart from Fintek to set the
> "speed mode" by default on this chip. It would be safer to default to
> full speed. OTOH the default speed target seems to be 4297 RPM if I
> read the datasheet correctly, which is high enough. If the fan speed
> reading isn't reliable (which admittedly is a bad thing to start with)
> I'd expect the chip to set the fan output to the max. Isn't it the case?
>
If it would be setting fan speed to full speed by default
I would not be sending these patches in the first place...
>> With, say, 2x10 rpm disks inside the case, I'd say the risk of
>> disks breaking down prematurely is very real. And I do not
>> want to take the responsibility of people reporting that "installing
>> Debian caused my hardware to break down"
>
> I see. But I can't think of a reason why this can't be simply solved in
> user-space, as Thecus is doing in their own product. This would be much
> easier than the kernel patches you've been sending.
Exactly what should Debian ship in /etc/sensors.conf to make
f75375s turn fans on N2100 systems but nowhere else? Simple
my ASS. Last I read about operating systems, the bloody point
of kernel was to abstract hardware differences.

Conveying platform_data from board configs to drivers
is a well established practice to solve platform specific quirks.

Not to mention that these kind of safety measures should be
done as early as possible. And due to the reasons I already
gave, changing bootloader is not really a option, the kernel
is the next best thing.

>>> Can I see a dump of the chip before the Linux driver is loaded?
>> I'll grab it, but see the first paragraph I wrote.
>
> I'm still interested.
>
> Thanks,
     0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f    0123456789abcdef
00: 01 03 82 40 5c 18 ff ff ff ff ff ff ff ff ff ff    ???@\?..........
10: d8 a8 d7 72 24 22 ff ff ff ff ff ff ff ff ff ff    ???r$"..........
20: ff 00 ff 00 ff 00 ff 00 3c 37 3c 37 ff ff ff ff    ........<7<7....
30: 00 00 00 00 00 f0 00 03 00 ff ff ff ff ff ff ff    .....?.?........
40: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff    ................
50: ff ff ff ff ff ff ff ff ff ff 03 06 15 19 34 ff    ..........????4.
60: 50 02 09 00 00 00 09 89 89 30 30 10 10 33 32 00    P??...???00??32.
70: ff ff ff ff ff ff 00 05 0a ff ff ff f5 00 ff ff    .......??...?...
80: ff ff ff ff ff ff 00 05 0a ff ff ff f5 00 ff ff    .......??...?...
90: 00 00 0d ff ff ff ff ff ff ff ff ff ff ff ff ff    ..?.............
a0: 3c 32 28 1e ff ff ff ff ff ff ff ff ff ff ff ff    <2(?............
b0: 3c 32 28 1e ff ff ff ff ff ff ff ff ff ff ff ff    <2(?............
c0: ae 04 ae 04 ff ff ff ff ff ff ff ff ff ff ff ff    ????............
d0: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff    ................
e0: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff    ................
f0: 80 80 00 00 00 03 5b 06 c0 00 00 ff ff ff ff 1d    ??...?[??......?




_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Mark M. Hoffman]

Hi:

* Lennert Buytenhek <buytenh@wantstofly.org> [2007-10-16 13:40:35 +0200]:
> On Tue, Oct 16, 2007 at 01:08:57PM +0300, Riku Voipio wrote:
> 
> > On thecus n2100, the bootloader does not setup fans to run. In order
> > to protect the user from frying their gear, start up fans on boot.
> > 
> > Depends on:
> > 
> > f75375s: Add new style bindings
> > f75375s: allow setting up fans with platform_data
> > 
> > patches sent to lm-sensors mailing list.
> > 
> > Signed-off-by: Riku Voipio <riku.voipio@movial.fi>
> 
> Acked-by: Lennert Buytenhek <buytenh@wantstofly.org>

Applied to hwmon-2.6.git/testing, thanks.

-- 
Mark M. Hoffman
mhoffman@lightlink.com


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] [PATCH] On n2100 systems,

[Jean Delvare]

Hi Riku,

On Wed, 24 Oct 2007 17:03:28 +0300, Riku Voipio wrote:
>      0  1  2  3  4  5  6  7  8  9  a  b  c  d  e  f    0123456789abcdef
> 00: 01 03 82 40 5c 18 ff ff ff ff ff ff ff ff ff ff    ???@\?..........
> 10: d8 a8 d7 72 24 22 ff ff ff ff ff ff ff ff ff ff    ???r$"..........
> 20: ff 00 ff 00 ff 00 ff 00 3c 37 3c 37 ff ff ff ff    ........<7<7....
> 30: 00 00 00 00 00 f0 00 03 00 ff ff ff ff ff ff ff    .....?.?........
> 40: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff    ................
> 50: ff ff ff ff ff ff ff ff ff ff 03 06 15 19 34 ff    ..........????4.
> 60: 50 02 09 00 00 00 09 89 89 30 30 10 10 33 32 00    P??...???00??32.
> 70: ff ff ff ff ff ff 00 05 0a ff ff ff f5 00 ff ff    .......??...?...
> 80: ff ff ff ff ff ff 00 05 0a ff ff ff f5 00 ff ff    .......??...?...
> 90: 00 00 0d ff ff ff ff ff ff ff ff ff ff ff ff ff    ..?.............
> a0: 3c 32 28 1e ff ff ff ff ff ff ff ff ff ff ff ff    <2(?............
> b0: 3c 32 28 1e ff ff ff ff ff ff ff ff ff ff ff ff    <2(?............
> c0: ae 04 ae 04 ff ff ff ff ff ff ff ff ff ff ff ff    ????............
> d0: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff    ................
> e0: ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff    ................
> f0: 80 80 00 00 00 03 5b 06 c0 00 00 ff ff ff ff 1d    ??...?[??......?

Register 0x60 has value 0x50, meaning that both fans are in manual
mode. This is _not_ the chip default according to the datasheet.
Registers 0x76 and 0x86 have value 0x00, which mean PWM duty cycle at
0% for both fans. Again this is _not_ the chip default according to the
datasheet.

So, assuming that the dump above was really taken right after boot
before the Linux kernel driver had a chance to change register values,
and also assuming that the datasheet provided by Fintek is correct,
this really means that Thecus are voluntarily stopping the fans at boot
time. That's a pretty stupid thing to do.

This doesn't imply whether we should or shouldn't start the fans at a
later point, of course. But at least it explains why the fans aren't
spinning at boot when the datasheet defaults suggested that they should.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors