[linux-dvb] Re: Experience building DVS dvb-kernel on 2.4.18 forbudget Nova-T

[Michael Hunold <hunold-ml@web.de>]

Hello Stephen,

> I pulled it from the CVS - yesterday's version (2003-01-26).
Ok, thanks for your feedback.

> Here is my experience:
>
> 1) 2.4.18 kernel doesn't export get_user_pages
>
>   I just modified and rebuilt my kernel with this symbol exported.
Ok. It seems that this is really distribution specific -- with the 
RedHat patched 2.4.18-3 kernel that came with my RedHat 7.3 I could 
compile the driver.

Anyway, with a recent and offical 2.4.20 kernel, "get_user_pages()" gets 
exported.

> 2) Had some problems with GPLONLY symbols in saa7146*.c and video-buf.c
>
>   I don't really understand this stuff properly - I just replaced the
>   EXPORT_SYMBOL_GPL with a regular EXPORT_SYMBOL
Hmm, I never encountered these problems.

> 3) "unlikely" was unknown in video-buf.c
>
>   Probably because my gcc is version 2.95.3.  I just defined
>   unlikely with "#define unlikely(x) (x)"
Is it really a compiler thing? I use
> gcc version 2.96 20000731 (Red Hat Linux 7.3 2.96-110)
and never got this error.

> 4) in compat.[hc], vmalloc_to_page included for all kernels <2.5.0.
>
>   My kernel, at least, does have this exported.  Perhaps that's not
>   standard.  In any event, I changed the compat.c and compat.h to
>   only include it for versions <2.4.18
These stuff has been re-introduced into compat.[ch] and the one who did 
that was not aware that it's exported for kernels > 2.4.18. I don't have 
any compile problems, though.

> 5) In video-buf.c, BUG_ON wasn't defined
>
>   This was only introduced in 2.4.20.  There is a define for it in
>   saa7146.h, but video-buf.c doesn't #include that.
>   I changed video-buf.c to include saa7146.h
"video-buf" is a helper module that's not part of the dvb/saa7146 
driver, so it should not include saa7146.h. Instead, it should include 
<linux/kernel.h> directly probably.

> With this lot adjusted, the driver built and showed signs of working.
> My problem then was that dvbstream would capture a few seconds of video
> and then data would stop.  In the syslog, the fidbirqs would stop too.
> This seemed to correspond to a call to budget_diseqc_ioctl - which was
> odd because I didn't see in dvbstream where such an ioctl was called.
Unfortunately, I cannot comment on all that follows.

> Thanks, hope this feedback is helpful.
> Steve
CU
Michael.




--
Info:
To unsubscribe send a mail to listar@linuxtv.org with "unsubscribe linux-dvb" as subject.