Re: [PATCH 2/3] net: irda: pxaficp_ir: convert to readl and writel

[Petr Cvek <petr.cvek@tul.cz>]

Dne 11.9.2015 v 08:18 Robert Jarzmik napsal(a):
> Petr Cvek <petr.cvek@tul.cz> writes:
> 
>>> Should have been posted to linux arm kernel mailing list, unless my mailer
>>> failed ...
>>>
>> Searching for:
>>
>> 	"ARM: pxa: add resources to pxaficp_ir"
>>
>> did not found anything, same was for "ficp" in the linux-arm-kernel/netdev/linux-kernel
>> mailing list archive.
> Ah ok, I'll resend it then.
> 
>>>> BTW This patch required update of my kernel repo. It seems that my:
>>>>
>>>> 	magician.c patches + ficp patch + new dma engine
>>>>
>>>> does not work for me at all. Kernel throws some panic about interrupts and then
>>>> it ends in an infinite stack dumping loop. Fault occurs before rootfs is
>>>> mounted, so probably around MMC init (with removed SD card it fails normally
>>>> with no rootfs found error).
>>> Could you send me (privately) the stack you're getting please. This is something
>>> I'd like to catch up early in the -rc releases.
>>
>> Well this will be problem as I cannot save anything to an SD card after and
>> during the failure.  Only viable interfaces would be earlycon on an infraport
>> or high speed camera on LCD :-).
> Ah just as on my mioa701. I ended up soldering a JTAG cable :)
Rather not :-D.

> 
>>> And it is true I have not tested the rootfs special case, where drivers are not
>>> yet initialized (and more specifically gpio and interrupt chip). Your backtrace
>>> should tell me if you fall into this category of issues ... but I digress, this
>>> has no link with pxaficp.
>>
>> Should I start new thread? (same bug can be present in the FICP too)
> Yes, this pxamci bothers me, it deserves a thread.

Will start soon.

>>
>> Anyway after mmc dma revert I was still not able to start FICP. There is an error:
>>
>> 	Unable to handle kernel paging request at virtual address 32e00004
>>
>> from pxa_irda_startup() and it seems it is caused by register definitions. For example:
>>
>> 	writel_relaxed((val), (irda)->stuart_base + (off));
>>
>> is called by
>>
>> 	stuart_writel(si, 0, STIER);
>>
>> but STIER is not just an offset, but full register address:
>> 	
>> 	__REG(0x40700004)
>>
>> So the definition should be changed, unless there is another patch I did not
>> received (in that case, send me full patchset again please) :-).
> Agreed, this is a bug in this patch. With this fix, is the pxaficp working or do
> you need a bit more time to experiment ?

I have tried with a nasty hack (use only lower part of address, it should equal with reg offset):
	#undef __REG
	-#define __REG(x) (x)
	+#define __REG(x) (x & 0xffff)

and it seems to work. The module inits and I am able to see IrDA traffic and ping other machine. FIR mode (mostly impacted by DMA) is still untested as magician unfortunately supports only SIR mode.

Petr