Re: [PATCH] serial: parisc: GSC: fix build when PCI_LBA is not set

[Helge Deller]

On 2/14/22 18:47, Randy Dunlap wrote:
> Hi Helge,
>
> On 2/14/22 04:24, Helge Deller wrote:
>> On 2/14/22 13:05, Helge Deller wrote:
>>> On 2/14/22 01:15, Randy Dunlap wrote:
>>>> Hi,
>>>>
>>>> On 2/13/22 14:15, Helge Deller wrote:
>>>>> On 2/13/22 22:07, Randy Dunlap wrote:
>>>>>>
>>>>>>
>>>>>> On 2/13/22 12:35, Helge Deller wrote:
>>>>>>> Hi Randy,
>>>>>>>
>>>>>>> On 2/13/22 20:39, Randy Dunlap wrote:
>>>>>>>> There is a build error when using a kernel .config file from
>>>>>>>> 'kernel test robot' for a different build problem:
>>>>>>>>
>>>>>>>> hppa64-linux-ld: drivers/tty/serial/8250/8250_gsc.o: in function `.LC3':
>>>>>>>> (.data.rel.ro+0x18): undefined reference to `iosapic_serial_irq'
>>>>>>>>
>>>>>>>> when:
>>>>>>>>   CONFIG_GSC=y
>>>>>>>>   CONFIG_SERIO_GSCPS2=y
>>>>>>>>   CONFIG_SERIAL_8250_GSC=y
>>>>>>>>   CONFIG_PCI is not set
>>>>>>>>     and hence PCI_LBA is not set.
>>>>>>>>   IOSAPIC depends on PCI_LBA, so IOSAPIC is not set/enabled.
>>>>>>>>
>>>>>>>> Making SERIAL_8250_GSC depend on PCI_LBA prevents the build error.
>>>>>>>
>>>>>>> It maybe makes the build error go away, but ...
>>>>>>>
>>>>>>>> Signed-off-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
>>>>>>>> Reported-by: kernel test robot <lkp@xxxxxxxxx>
>>>>>>>> Cc: "James E.J. Bottomley" <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx>
>>>>>>>> Cc: Helge Deller <deller@xxxxxx>
>>>>>>>> Cc: linux-parisc@xxxxxxxxxxxxxxx
>>>>>>>> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
>>>>>>>> Cc: linux-serial@xxxxxxxxxxxxxxx
>>>>>>>> Cc: Jiri Slaby <jirislaby@xxxxxxxxxx>
>>>>>>>> Cc: Johan Hovold <johan@xxxxxxxxxx>
>>>>>>>> ---
>>>>>>>>  drivers/tty/serial/8250/Kconfig |    2 +-
>>>>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>>>>
>>>>>>>> --- linux-next-20220211.orig/drivers/tty/serial/8250/Kconfig
>>>>>>>> +++ linux-next-20220211/drivers/tty/serial/8250/Kconfig
>>>>>>>> @@ -118,7 +118,7 @@ config SERIAL_8250_CONSOLE
>>>>>>>>
>>>>>>>>  config SERIAL_8250_GSC
>>>>>>>>  	tristate
>>>>>>>> -	depends on SERIAL_8250 && GSC
>>>>>>>> +	depends on SERIAL_8250 && GSC && PCI_LBA
>>>>>>>>  	default SERIAL_8250
>>>>>>>
>>>>>>> The serial device is on the GSC bus, so if you make it
>>>>>>> dependend on the PCI bus it will not be useable on machines
>>>>>>> which only have a GSC bus...
>>>>>>>
>>>>>>> We need another patch.
>>>>>>> Do you have a link to the build error?
>>>>>>
>>>>>>
>>>>>> No, it's from the other build error that you just replied to,
>>>>>> where the incorrect compiler was used.
>>>>>>
>>>>>> I'll recheck it and reconsider what to do, if anything.
>>>>>
>>>>> Ok, thank you!
>>>>
>>>> I dunno what to do. This:
>>>>
>>>> #ifdef CONFIG_64BIT
>>>> 	if (!dev->irq && (dev->id.sversion == 0xad))
>>>> 		dev->irq = iosapic_serial_irq(dev);
>>>> #endif
>>>>
>>>> makes it look like 64BIT requires IOSAPIC (hence PCI_LBA).
>>>
>>> Although I think all 64bit machines have a PCI bus, the better
>>> fix is that the driver should only call iosapic_serial_irq(dev)
>>> if CONFIG_IOSAPIC is set. This patch fixes the build:
>>>
>>> -#ifdef CONFIG_64BIT
>>> +#ifdef CONFIG_IOSAPIC
>>>         if (!dev->irq && (dev->id.sversion == 0xad))
>>>                 dev->irq = iosapic_serial_irq(dev);
>>>  #endif
>>
>> That was not fully correct.
>> It needs to be:
>>
>> #if defined(CONFIG_64BIT) && defined(CONFIG_IOSAPIC)
>>
>> Otherwise you'll get an undefined reference in the 32-bit build.
>
> Sure, I can send such a patch.

Thanks!
I see you sent it in a seperate mail.
I can take it through the parisc tree.

> I would have used a bigger hammer and done something like
>
> 	depends on IOSAPIC if 64BIT
>
> Just for info, how would dev->irq be set for CONFIG_64BIT
> when CONFIG_IOSAPIC is not set?

All found devices in the parisc system are in a device table.
The serial driver scans those against the ones listed in
serial_tbl[] and lasi_tbl[], and hand over the parisc_device
pointer which usually has the irq set.
In case the device is connected via iosapic and hasn't an irq
already set, iosapic_serial_irq() is called to find the irq.
The
 	depends on IOSAPIC if 64BIT
would have dropped *all* serial ports on 64bit kernels, even
those which are on the GSC bus.

Helge