Re: [lm-sensors] 2.6.24-rc4 hwmon it87 probe fails

Re: [lm-sensors] 2.6.24-rc4 hwmon it87 probe fails

[Robert Hancock]

Carlos Corbacho wrote:
> On Thursday 20 December 2007 00:20:21 Bjorn Helgaas wrote:
>> I suspect the manufacturers would say "Oh, the sensors?  The BIOS
>> isn't broken, you're just supposed to use WMI or some (undocumented)
>> ACPI device to get at those."
> 
> It's quite possible - can we have DSDTs for the boards in question so we can 
> quickly check if this is a possibility? (Basically, to see if they have 
> PNP0C14 devices - if they don't, then I'm afraid it's nothing to do with 
> WMI).
> 
> -Carlos

It's quite possible that the BIOS accesses the device either from ACPI 
AML or possibly even from SMI. In that case it would be quite reasonable 
for the BIOS to reserve that region to prevent another driver from 
loading and trying to take conflicting control of the device. One has to 
be careful before assuming that any such reservation is bogus.

-- 
Robert Hancock      Saskatoon, SK, Canada
To email, remove "nospam" from hancockr@nospamshaw.ca
Home Page: http://www.roberthancock.com/

Re: 2.6.24-rc4 hwmon it87 probe fails

[Jean Delvare]

Hi Bjorn,

Le 21/12/2007, "Bjorn Helgaas" <bjorn.helgaas@hp.com> a écrit:

>On Tuesday 18 December 2007 10:59:18 am Jean Delvare wrote:
>> My initial idea was to identify the faulty motherboard using DMI and to
>> force pnpacpi=off on the faulty motherboards. If this is considered too
>> aggressive, maybe we can just reject resource declarations that
>> intersect (but don't match) 0x290-0x297 for these motherboards. Either
>> way, we have to do something, and we have to do it quickly. 2.6.24
>> final isn't too far away, and more importantly, the patch that revealed
>> the problem has been backported to 2.6.23.10 so people are experiencing
>> regressions already.
>
>What do you think of something like the following patch?  If we do
>this, I don't think we'd need to force pnpacpi=off or change the
>way PNP reserves resources.
>
>I'll be on vacation until about January 2, so I won't be able to
>do much with this until then.
>
>
>
>[patch] it87: request only Environment Controller ports
>
>The IT8705F and related parts are Super I/O controllers that contain
>many separate devices.
>
>Some BIOSes describe IT8705F I/O port usage under a motherboard device
>(PNP0C02) with overlapping regions, e.g., 0x290-0x29f and 0x290-0x294.
>
>The it87 driver supports only the Environment Controller, which requires
>only two ISA ports, but it used to request an eight-port range.  If that
>range exceeds a range reported by the BIOS, as 0x290-0x297 would, the
>request fails, and the it87 driver cannot claim the device.
>
>This patch makes the it87 driver request only the two ports used for the
>Environment Controller device.

The problem is that the IT87xxF chips do decode 4 ports (recent chips,
0x294-0x297) or 8 ports (older chips, 0x290-0x297), not 2 as the
datasheets say. The ITE Super-I/O chips differ from the Winbond
Super-I/O chips in this respect. The it87 driver only needs to access
ports 0x295 and 0x296, true, but the device itself decodes more
addresses than that. So, with this proposed patch, ports which are busy
will be shown as being free. This pretty much voids the point of
resource declarations, doesn't it? This might not cause too much
trouble in practice, but to me this still looks like the wrong way to go.

The resource declarations made by these motherboard BIOSes are totally
bogus. 0x290-0x29f is much larger than what the chip decodes.
0x290-0x294 is a subset that doesn't make any sense. The GA-K8N Ultra 9
is even funnier with 0x295-0x314, which again doesn't correspond to
anything real. All these happen to not intersect with 0x295-0x296 but I
wouldn't count on it. If Gigabyte (and possibly others) care that
little about these declarations, pretty much anything could be seen. So
while your proposed workaround happens to fix the problem at hand, it is
not really correct technically, and could break again soon.

I'd rather fix the problem at its source, or, as fixing it as the source
isn't very realistic in this case, as near of the source as possible.
That would mean DMI-based quirks for the affected motherboards, that
would discard or adjust the bogus resource declarations. I have not yet
looked at what the code would look like, as I am not familiar with the
pnp system driver, I don't know what it would take to implement my
proposal.

I also don't fully understand what pnpacpi is useful for. I have heard
about PCI drivers that might request resources that the motherboard
doesn't want them to touch, but I don't know the details, I also
don't know if it is a theoretical issue or if it really happens on some
systems, and I don't know if there are other uses for pnpacpi. I'm
asking because I want to know if forcing pnpacpi=off on the faulty
motherboards would be good enough or if we really need finer-grained
quirks (assuming that we go the DMI-based quirks route at all.) The
former would obviously be easier.

>
>Systems where this problem has been reported:
>    Gigabyte GA-K8N Ultra 9
>    Gigabyte M56S-S3
>    Gigabyte GA-965G-DS3
> (...)

Thanks,
--
Jean Delvare

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

Re: 2.6.24-rc4 hwmon it87 probe fails

[Jean Delvare]

Hi Bjorn,

Le 23/12/2007, "Bjorn Helgaas" <bjorn.helgaas@hp.com> a écrit:
>On Sunday 23 December 2007 2:28:05 am Jean Delvare wrote:
>> The problem is that the it87 driver is used on a variety of motherboards,
>> some where the hardware monitoring device I/O ports are reserved by the
>> BIOS, some where they aren't. How am I supposed to deal with this?
>
>I assume you mean some boards describe those ports in ACPI, and some
>don't?

Yes, that's what I meant.

>(...) I think the ideal solution would be to figure out how the
>BIOS writers intended the device to be used, and do that.  But the
>documentation may be lacking, and it degenerates into an OEM-specific
>mess.  Second-best seems like a PNP quirk that pokes enough to determine
>that a Super I/O device is present, and then reserves resources for
>it.  Then we avoid the collision even if it87 isn't present.

I'm not sure what you mean here exactly. We have two different cases:
either the ACPI code did declare the PNP device, or it did not. How do
you unify both? And what "collision" are you talking about?

>> PCI is an entirely different beast. With PCI you know the PCI device ID
>> that is associated with the resources, and for a given device, the
>> resources are always declared (if standard BARs are used) or never
>> declared; there's no "may be". So it's much easier to handle the
>> resources properly.
>
>That's the way PNP is supposed to work, too (more about this below).

Not really. PNP devices may or may not be declared. PCI devices are
always declared. This is a fundamental difference. I'm also not sure
about the PNP IDs, see below.

>> That's not really safe, no. They may overlap with other device resources
>> and prevent those drivers from loading.
>
>True.  If that happens, I think we definitely need some kind of DMI-
>based quirk to fix the resources.  My points are (a) the quirk needs
>to be at the PNP level; it can't be in a driver, and (b) I'm lazy and
>would wait until a problem happens before implementing it, because I
>know so little about PNP and the specific board, and by waiting, there's
>a chance I'll learn enough to avoid a mistake :-)

But we _do_ have a problem to solve! That's what started this thread.
And we need to solve it before 2.6.24 is released. I'd rather (try to)
do the right thing now rather than going in one direction today and
doing something different next month.

I do agree that the quirks should be done at the PNP level and not in the
it87 driver.

>> Another problem is how do I connect this specific I/O port range of the
>> PNP0C02 device with the it87 driver? I am by far no PNP expert but it
>> looks to me like this PNP device covers more than one actual device, and
>> I/O port ranges do not have labels nor identifiers that would let me
>> find the one that corresponds to the IT87xxF device (if it exists at
>> all.)
>
>The general rule is that you have a PNP device with an identifier
>like PNP0500 that means "16550 UART" and some resources underneath
>it.  The PNP ID ("PNP0500") tells the OS which driver to bind to the
>device, and the resources tell the driver where the device is.  The
>serial driver in drivers/serial/8250_pnp.c is a good example of the
>"normal" way PNP drivers work.
>
>it87 doesn't fit the model very well because usually the BIOS
>doesn't describe the device explicitly.  I think the expectation
>is that the OS will get sensor readings some other way, such as
>by evaluating ACPI "_TMP" methods, or by using some OEM-specific
>ACPI device.

If the IT87xxF device is declared, will it always the same PNP ID on
every system? I fear that the answer is no, in which case I see no
simple way to implement the it87 driver as a PNP driver.

>It's very irritating when ACPI is used to take some device that
>would otherwise be nicely generic and machine-independent, and
>wrap it up into some abstract device that requires an OEM-specific
>driver, but I think that's what's happening here.  I suspect it
>has to do with the fact that the BIOS wants to retain some
>control over the device so it can deal with things like thermal
>emergencies, even if the OS driver is missing or broken.

Yes, very irritating. Especially when the ACPI way is not only
system-specific, but also very poor in terms of features compared to
what a native driver can do. This is the main reason why we still write
native drivers for hardware monitoring drivers despite the conflicts
with ACPI that arise on some systems.

--
Jean Delvare
-
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: 2.6.24-rc4 hwmon it87 probe fails

[Bjorn Helgaas]

On Tuesday 25 December 2007 02:31:26 pm Jean Delvare wrote:
> Le 23/12/2007, "Bjorn Helgaas" <bjorn.helgaas@hp.com> a écrit:
> >On Sunday 23 December 2007 2:28:05 am Jean Delvare wrote:
> >> The problem is that the it87 driver is used on a variety of motherboards,
> >> some where the hardware monitoring device I/O ports are reserved by the
> >> BIOS, some where they aren't. How am I supposed to deal with this?
> >
> >I assume you mean some boards describe those ports in ACPI, and some
> >don't?
> 
> Yes, that's what I meant.
> 
> >(...) I think the ideal solution would be to figure out how the
> >BIOS writers intended the device to be used, and do that.  But the
> >documentation may be lacking, and it degenerates into an OEM-specific
> >mess.  Second-best seems like a PNP quirk that pokes enough to determine
> >that a Super I/O device is present, and then reserves resources for
> >it.  Then we avoid the collision even if it87 isn't present.
> 
> I'm not sure what you mean here exactly. We have two different cases:
> either the ACPI code did declare the PNP device, or it did not. How do
> you unify both?

Even if the BIOS does not declare an IT87xxF as an independent device,
there may be AML that uses the chip internally.  For example, the
BIOS could declare a thermal zone with a _TMP method, and the _TMP
method could use the IT87xxF to read the temperature.  In that case,
the BIOS would still need to declare the IT87xxF resources so the OS
doesn't place another device on top of it.  An easy way to do this
is with a PNP0C02 "motherboard" device, which just says "these
resources are in use, but the OS needn't attach a driver to this
device."

> And what "collision" are you talking about? 

I meant the problem where we place two devices at the same address,
causing one to stop working.  I wrote "even if it87 isn't present,"
which is ambiguous -- I meant that we need something that works even
when the it87 _driver_ isn't present.

> >> PCI is an entirely different beast. With PCI you know the PCI device ID
> >> that is associated with the resources, and for a given device, the
> >> resources are always declared (if standard BARs are used) or never
> >> declared; there's no "may be". So it's much easier to handle the
> >> resources properly.
> >
> >That's the way PNP is supposed to work, too (more about this below).
> 
> Not really. PNP devices may or may not be declared. PCI devices are
> always declared. This is a fundamental difference. I'm also not sure
> about the PNP IDs, see below.

Well, it's true that PNP does not need to describe ALL devices in the
system.  However, it should describe all devices to which the OS should
bind a driver.  The BIOS writer has the discretion to hide devices from
the OS by omitting them from the ACPI namespace.  If he does that, his
assumption is that the OS will ignore the device (but of course, he's
thinking about a Plug-and-Play-compliant OS like Windows).

> >> That's not really safe, no. They may overlap with other device resources
> >> and prevent those drivers from loading.
> >
> >True.  If that happens, I think we definitely need some kind of DMI-
> >based quirk to fix the resources.  My points are (a) the quirk needs
> >to be at the PNP level; it can't be in a driver, and (b) I'm lazy and
> >would wait until a problem happens before implementing it, because I
> >know so little about PNP and the specific board, and by waiting, there's
> >a chance I'll learn enough to avoid a mistake :-)
> 
> But we _do_ have a problem to solve! That's what started this thread.
> And we need to solve it before 2.6.24 is released. I'd rather (try to)
> do the right thing now rather than going in one direction today and
> doing something different next month.

I think my patch to request only the ports actually used by the it87
driver is sufficient to solve the current problem.  If we find that
a BIOS neglects to reserve some of the ports decoded by the IT87xxF,
we can add a PNP quirk to handle that.  But I'm not aware of any
problems like that yet.

> >it87 doesn't fit the model very well because usually the BIOS
> >doesn't describe the device explicitly.  I think the expectation
> >is that the OS will get sensor readings some other way, such as
> >by evaluating ACPI "_TMP" methods, or by using some OEM-specific
> >ACPI device.
> 
> If the IT87xxF device is declared, will it always the same PNP ID on
> every system? I fear that the answer is no, in which case I see no
> simple way to implement the it87 driver as a PNP driver.

PNP drivers can claim multiple PNP IDs, e.g., drivers/serial/8250_pnp.c
recognizes many IDs for 16550-compatible UARTs.  But I doubt that
any BIOS actually describes an IT87xxF explicitly.

Bjorn
-
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: 2.6.24-rc4 hwmon it87 probe fails

[Jean Delvare]

Hi Bjorn,

Sorry for the late answer.

On Wed, 2 Jan 2008 11:30:55 -0700, Bjorn Helgaas wrote:
> Even if the BIOS does not declare an IT87xxF as an independent device,
> there may be AML that uses the chip internally.  For example, the
> BIOS could declare a thermal zone with a _TMP method, and the _TMP
> method could use the IT87xxF to read the temperature.  In that case,
> the BIOS would still need to declare the IT87xxF resources so the OS
> doesn't place another device on top of it.

Thomas Renninger's patch set should handle this case. It's in -mm at the
moment and I guess we'll merge most of it in 2.6.25.

> (...) An easy way to do this
> is with a PNP0C02 "motherboard" device, which just says "these
> resources are in use, but the OS needn't attach a driver to this
> device."

How do I see if a given motherboard does this or not? For example I
have a motherboard here that says:

0290-029f : pnp 00:04

How do I know if this is a "PNP0C02 motherboard device" I am supposed
not to touch, or a device that I can attach a native hwmon driver to?
And more importantly, how does the it87 driver (or any other hwmon
driver) know about it? I'm all for updating the hwmon drivers to
cooperate better with PNP in order to prevent clashes between the BIOS
and the OS, but I just don't know how I am supposed to do that.

> > And what "collision" are you talking about? 
> 
> I meant the problem where we place two devices at the same address,
> causing one to stop working.  I wrote "even if it87 isn't present,"
> which is ambiguous -- I meant that we need something that works even
> when the it87 _driver_ isn't present.

I agree then. I don't think this has much to do with PNP though. What
this is calling for is a Super-I/O subsystem that detects the Super-I/O
and instantiate devices out of it. That way the drivers for these
devices (in particular hardware monitoring drivers, watchdog drivers,
some parport drivers...) would no longer need to instantiate their
devices themselves, thus matching the device driver model much better.
I've seen patches float around but so far I could never find the time
to look into them (and I fear it's unlikely to change anytime soon.)

> (...)
> Well, it's true that PNP does not need to describe ALL devices in the
> system.  However, it should describe all devices to which the OS should
> bind a driver.  The BIOS writer has the discretion to hide devices from
> the OS by omitting them from the ACPI namespace.  If he does that, his
> assumption is that the OS will ignore the device (but of course, he's
> thinking about a Plug-and-Play-compliant OS like Windows).

Even under Windows, I pretty much doubt that any monitoring application
cares about what the BIOS says. Motherboard vendor applications have
enough information about each board model to know what they can do or
not without asking the BIOS. Third-party applications probably just
poke at known I/O locations (pretty much like the Linux hwmon drivers
do) and hope for the best.

> (...)
> I think my patch to request only the ports actually used by the it87
> driver is sufficient to solve the current problem.  If we find that
> a BIOS neglects to reserve some of the ports decoded by the IT87xxF,
> we can add a PNP quirk to handle that.  But I'm not aware of any
> problems like that yet.

I'm not aware of such a problem either.

-- 
Jean Delvare

Re: 2.6.24-rc4 hwmon it87 probe fails

[Jean Delvare]

Hi Robert,

On Wed, 19 Dec 2007 19:09:54 -0600, Robert Hancock wrote:
> It's quite possible that the BIOS accesses the device either from ACPI 
> AML or possibly even from SMI. In that case it would be quite reasonable 
> for the BIOS to reserve that region to prevent another driver from 
> loading and trying to take conflicting control of the device. One has to 
> be careful before assuming that any such reservation is bogus.

Again I am all for honoring such BIOS requests so as to prevent
conflicts between ACPI or SMI and native drivers. The problem is that
no two BIOS out there do the same in this respect. I couldn't see any
correlation between machines declaring their hwmon device in PNP and
machines where ACPI or SMI access the device in question. Many boards
declare their device and seemingly never touch it so it's fine for
Linux to drive them. Some boards no not declare the devices but still
access them in our back.

Thomas' patches should deal with the ACPI AML case in most cases, but
not with SMI.

So either the PNP code in Linux isn't exporting enough details to
differentiate, or even the PNP code has no way to tell these cases
apart. In the latter case there's not much we can do. In the former
case, let's have the PNP code export the information so that hwmon
drivers can decide whether they should bind to the devices or not by
default.

-- 
Jean Delvare