[lm-sensors] [PATCH] remove the fix_hstcfg from i2c-piix4

[lm-sensors] [PATCH] remove the fix_hstcfg from i2c-piix4

[Rudolf Marek]

Hello,

This patch removes the fix_hstcfg option from the driver and related SMBus Interrupt Select register
magic because now we know what are valid values for this register. This patch updates the documentation
and adds new IRQ mode check so we are sure not to miss any new "unusual" value.

The PCI quirk for users of fix_hstcfg was not developed because the chipset lacks of subsystem ID registers
and DMI is stated "To be filled". Impact to existing systems is minimal because the problem showed up
on motherboards like 10 years back. On the other hand users of newer Serverworks and HT1000 systems won't
be misleaded by the message suggesting to try the fix_hstcfg any more.

Signed-off-by: Rudolf Marek <r.marek at sh.cvut.cz>

Regards
Rudolf


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_remove_fixhstcfg
Url: http://lists.lm-sensors.org/pipermail/lm-sensors/attachments/20060402/9d3d753a/patch_remove_fixhstcfg.pl

[lm-sensors] [PATCH] remove the fix_hstcfg from i2c-piix4

[Jean Delvare]

Hi Ruik,

> This patch removes the fix_hstcfg option from the driver and related SMBus Interrupt Select register
> magic because now we know what are valid values for this register. This patch updates the documentation
> and adds new IRQ mode check so we are sure not to miss any new "unusual" value.
> 
> The PCI quirk for users of fix_hstcfg was not developed because the chipset lacks of subsystem ID registers
> and DMI is stated "To be filled". Impact to existing systems is minimal because the problem showed up
> on motherboards like 10 years back. On the other hand users of newer Serverworks and HT1000 systems won't
> be misleaded by the message suggesting to try the fix_hstcfg any more.

Agreed.

My comments on the patch:

> +If you have some rather strange SMBus problems, you may need to change the
> +SMBus Interrupt Select register as it is known solution for Force CPCI735
> +motherboard or other OSB4 based systems. Please contact us for further
> +assistance (lm-sensors at lm-sensors.org).

Oh, no. We don't want more support requests on this. Instead, please
explain how to get the register value using lspci and how to write back
the modified value using setpci.

Besides, are you sure that other OSB4 based systems did need this? I've
searched all my mails, and couldn't find anyone reporting success after
using fix_hstcfg, except Tom.

> -	if ((temp & 0x0E) = 8)
> +	if (((temp & 0x0E) = 8) || ((temp & 0x0E) = 2))

I think I understand that 8 is OK for non-Serverworks one and 2 is OK
for Serverworks one, right? So maybe we can improve this check and do:

if ((temp & 0x0E) = (id->vendor = PCI_VENDOR_ID_SERVERWORKS ? 2 : 8))

What do you think?

Thanks,
-- 
Jean Delvare

[lm-sensors] [PATCH] remove the fix_hstcfg from i2c-piix4

[Rudolf Marek]

Hi Jean
>>+If you have some rather strange SMBus problems, you may need to change the
>>+SMBus Interrupt Select register as it is known solution for Force CPCI735
>>+motherboard or other OSB4 based systems. Please contact us for further
>>+assistance (lm-sensors at lm-sensors.org).
> 
> 
> Oh, no. We don't want more support requests on this. Instead, please
> explain how to get the register value using lspci and how to write back
> the modified value using setpci.

Well I think there will be requests like same probability that you capture your
very own neutrino :)

Anyway I will prepare the DIY guide.

> 
> Besides, are you sure that other OSB4 based systems did need this? I've
> searched all my mails, and couldn't find anyone reporting success after
> using fix_hstcfg, except Tom.
> 

The old text suggest this. All I think is that the bios left the controller in
interrupt mode and I think our driver assumes always the SMI mode.


> 
>>-	if ((temp & 0x0E) = 8)
>>+	if (((temp & 0x0E) = 8) || ((temp & 0x0E) = 2))
> 
> 
> I think I understand that 8 is OK for non-Serverworks one and 2 is OK
> for Serverworks one, right? So maybe we can improve this check and do:
> 
> if ((temp & 0x0E) = (id->vendor = PCI_VENDOR_ID_SERVERWORKS ? 2 : 8))

I thought about this but for the reason that serverworks was bought by broadcom
and we have ATI and others there so best sounds to me to make it more global.

Regards
Rudolf

[lm-sensors] [PATCH] remove the fix_hstcfg from i2c-piix4 -

[Rudolf Marek]

Hello,

This patch removes the fix_hstcfg option from the driver and related SMBus
Interrupt Select register
magic because now we know what are valid values for this register. This patch
updates the documentation
and adds new IRQ mode check so we are sure not to miss any new "unusual" value.

The PCI quirk for users of fix_hstcfg was not developed because the chipset
lacks of subsystem ID registers
and DMI is stated "To be filled". Impact to existing systems is minimal because
the problem showed up
on motherboards like 10 years back. On the other hand users of newer Serverworks
and HT1000 systems won't
be misleaded by the message suggesting to try the fix_hstcfg any more.

Signed-off-by: Rudolf Marek <r.marek at sh.cvut.cz>

Regards
Rudolf
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: patch_remove_fixhstcfg2
Url: http://lists.lm-sensors.org/pipermail/lm-sensors/attachments/20060402/957ccd76/patch_remove_fixhstcfg2.pl

[lm-sensors] [PATCH] remove the fix_hstcfg from i2c-piix4 -

[Jean Delvare]

Hi Rudolf,

> This patch removes the fix_hstcfg option from the driver and related SMBus
> Interrupt Select register magic because now we know what are valid values
> for this register. This patch updates the documentation and adds new IRQ
> mode check so we are sure not to miss any new "unusual" value.
> 
> The PCI quirk for users of fix_hstcfg was not developed because the chipset
> lacks of subsystem ID registers and DMI is stated "To be filled". Impact to
> existing systems is minimal because the problem showed up on motherboards
> like 10 years back. On the other hand users of newer Serverworks and HT1000
> systems won't be misleaded by the message suggesting to try the fix_hstcfg
> any more.

Applied, thanks. I only changed the sample lspci line to use a real
OSB4 example (I have one.)

-- 
Jean Delvare