2008/10/6 Balaji Rao <[EMAIL PROTECTED]>: > On Sat, 27 Sep 2008 22:25:35 +0200 > "andrzej zaborowski" <[EMAIL PROTECTED]> wrote: > >> 2008/9/27 Balaji Rao <[EMAIL PROTECTED]>: >> > On Friday 26 September 2008 11:55:57 am Werner Almesberger wrote: >> > >> > <snip> >> > >> >> >> >> These are quite drastic changes. The PCF50633 driver has grown into >> >> being the "GTA02 schematics sheet 3" driver over time, so there's >> >> now quite a bit to untangle. >> >> >> > >> > Yes, one thing I noticed was that the pcf50633 driver works only >> > with one instance of the hardware present. I was working to remove >> > that limitation. I've used i2c_board_info from within mach-gta02.c >> > and changed the pcf50633 i2c_client driver to "new style". >> > >> > I'll begin looking into the other problems you mentioned. >> > >> > One problem now is that I'm not able to build stable-upstream. I see >> > a lot of errors and warnings related to pwm and fiq (Btw, what is >> > pwm and fiq ? - I'm not able to find any decent explanation >> > anywhere). Here are the build messages. Can anyone tell me what's >> > wrong ? One thing I've noticed is that the include/asm-arm/ >> > directory no longer contains machine specific includes. I believe >> > they have been moved into the arch/asm/arm/mach-xxxx/include >> > directory ? Is this the cause of >> >> I built last week's stable-tracking, there are various issues with it >> but it boots. I had to do the following manually: >> >> ln -s arch-s3c2410 include/asm-arm/arch >> >> Normally this was done by Makefile, I don't yet know why that changed. >> Beside the mtdblock issues I see there's also a funny problem with >> time running ~6x slower on my gta02. I already checked that it's not >> related to the dynticks patch. > > If I'm not wrong, by the "dynticks" patch you talk about your patch that > uses CONFIG_NO_IDLE_HZ. > > When I was going through that code today, I saw that CONFIG_NO_IDLE_HZ > is not there in any Kconfig file. So I was wondering if your code in > stable-tracking was being compiled at all.. Please clarify.
Right... it turns out CONFIG_NO_IDLE_HZ was completely removed from the upstream kernels at 2.6.21, but openmoko still used it, so I added dynticks support using CONFIG_NO_IDLE_HZ... Upstream the whole sys timer framework was replaced by what now is clocksources and clockevents frameworks and seemingly s3c24xx was the last architecture to not be converted yet. I submitted a patch today that attempts this: http://marc.info/?l=linux-arm-kernel&m=122321495315232&w=2 Hopefully we won't need our timer code anymore after this. Cheers
