Re: Problem with VMAP_STACK=y

[Andy Lutomirski]

On Wed, Oct 5, 2016 at 9:45 AM, JÃrg Otte <jrg.otte@xxxxxxxxx> wrote:
> 2016-10-05 17:53 GMT+02:00 Andy Lutomirski <luto@xxxxxxxxxxxxxx>:
>> On Wed, Oct 5, 2016 at 8:21 AM, JÃrg Otte <jrg.otte@xxxxxxxxx> wrote:
>>> 2016-10-05 11:04 GMT+02:00 Mauro Carvalho Chehab <mchehab@xxxxxxxxxxxxxxxx>:
>>>> Em Wed, 5 Oct 2016 09:50:42 +0200 (CEST)
>>>> Jiri Kosina <jikos@xxxxxxxxxx> escreveu:
>>>>
>>>>> On Wed, 5 Oct 2016, Patrick Boettcher wrote:
>>>>>
>>>>> > > > Thanks for the quick response.
>>>>> > > > Drivers are:
>>>>> > > > dvb_core, dvb_usb, dbv_usb_cynergyT2
>>>>> > >
>>>>> > > This dbv_usb_cynergyT2 is not from Linus' tree, is it? I don't seem
>>>>> > > to be able to find it, and the only google hit I am getting is your
>>>>> > > very mail to LKML :)
>>>>> >
>>>>> > It's a typo, it should say dvb_usb_cinergyT2.
>>>>>
>>>>> Ah, thanks. Same issues there in
>>>>>
>>>>>       cinergyt2_frontend_attach()
>>>>>       cinergyt2_rc_query()
>>>>>
>>>>> I think this would require more in-depth review of all the media drivers
>>>>> and having all this fixed for 4.9. It should be pretty straightforward;
>>>>> all the instances I've seen so far should be just straightforward
>>>>> conversion to kmalloc() + kfree(), as the buffer is not being embedded in
>>>>> any structure etc.
>>>>
>>>> What we're doing on most cases is to put a buffer (usually with 80
>>>> chars for USB drivers) inside the "state" struct (on DVB drivers),
>>>> in order to avoid doing kmalloc/kfree all the times. One such patch is
>>>> changeset c4a98793a63c4.
>>>>
>>>> I'm enclosing a non-tested patch fixing it for the cinergyT2-core.c
>>>> driver.
>>>>
>>>> Thanks,
>>>> Mauro
>>>>
>>>> [PATCH] cinergyT2-core: don't do DMA on stack
>>>>
>>>
>>> Tried the cinergyT2 patch. No success. When I select a TV channel
>>> just nothing happens. There are no error messages.
>>
>> Could you try compiling with CONFIG_DEBUG_VIRTUAL=y and seeing if you
>> get a nice error message?
>>
>> --Andy
>
> Done. Still no error messages in dmesg and syslog.
>
> DVB adapter signals it is tuned to the channel.
> For me it looks like there`s no data reaching the application
> (similar to if I forget to connect an antenna).

I'm surprised.  CONFIG_DEBUG_VIRTUAL=y really ought to have caught the
problem, whatever it is.  You could try CONFIG_DEBUG_SG as well, but I
admit I'm grasping at straws there.  Maybe the DVB people have a
better idea as to what's going on here.

It's plausible that the patch you're testing got rid of the DMA on the
stack but is otherwise buggy.

--Andy