Messages in this thread | | | Date | Sun, 23 Mar 2008 07:35:24 +0100 | From | Ingo Molnar <> | Subject | Re: ohci1394 problem (MMIO broken) (was 2.6.25-rc6-git6: Reported regressions from 2.6.24) |
| |
* Stefan Richter <stefanr@s5r6.in-berlin.de> wrote:
> On my x86-32 test system, none of the devices gets an MMIO region > above the 4G map. (Reconfigured with CONFIG_RESOURCES_64BIT=y while > having CONFIG_HIGHMEM4G=y, like in Thomas M's .config.)
64-bit ioremap on x86 never worked AFAIK, the whole API is using 32-bit physical addresses which is limited to 4GB. But ... this bug does have the feeling of the wrong IO side physical memory being mapped. (0xff data means there's no device there - typical of non-existent RAM/IO)
could you please pull the x86/latest tree to do a kernel pagetable dump for us? You can pull it via the instructions here:
http://people.redhat.com/mingo/x86.git/README
then enable CONFIG_DEBUG_KERNEL=y and CONFIG_X86_PTDUMP=y boot the kernel so that ohci1394 doesnt work. Then dump the pagetables and other hw info via:
http://people.redhat.com/mingo/cfs-scheduler/tools/cfs-debug-info.sh
send us the cfs-debug-info file that gets generated. We'll be able to tell from that whether your ioremap's are OK. (x86/latest will also printk all ioremap()s and results that are done)
thanks,
Ingo
| |