I did some tests with 2.6.16-rc5 on a number of platforms and here's
a status report.
Problems first. ;-)
- The iomap patch is not in the mips-linux tree yet. Needs support for
multiple PCI busses (as well for non-PCI busses?), see discussion at
- >2 min delay probing for hard drives. The following patch still
needs to be applied by Ralf:
- The IDE/network interaction crash has been greatly improved by
but is not 100% gone. Nobody on netdev commented on the patch yet.
- Possible corruption with USB (regression from 2.4) but I have
not had a chance to reproduce this yet; possibly related to
64-bit. Will ask bug reporter to test with a 32-bit kernel.
64-bit has a number of problems. Peter Horton says "What surprises me
is that all the problems appear to be timing related rather than 32/64
bit portability issues. Maybe this is 'cos the RM5231 has such small
caches; doubling the size of all the kernel's pointers probably impacts
us badly and slows everything down."
- Tulip driver thinks MAC address on Qube 2700 is "ff:ff:ff:ff:ff:ff"
- Tulip driver needs fix from Grant Grundler that has been nacked by
- PCI audio with ALSA doesn't work. In the past it failed with:
now we get:
- IP22 doesn't shutdown due to a oops in the serial driver. Patch at
- IP22: Indigo2 with > 256 MB fails to boot (regression from 2.4;
while 2.4 would only see 256 MB, it would at least boot)
- IP22: VINO support is broken on 64-bit:
- IP32: Apparently O2 with 384 MB RAM doesn't load, see
Kumba said IP32 works "only, it seems, on multiples like 64, 128, 256,
and 512, based on issues we've seen gentoo-side"
- IP32: FB doesn't work when you use > 4 MB RAM for it.
- The duart driver for SWARM is not in mainline anymore; a regression
- Apparently USB hid is broken (says p2); a regression from 2.4.
- There are a number of time related problems; patches available:
- gettimeofday jumps backwards then forwards (generic plus SB1250)
- fix lost ticks on SB1250
- bcm1480 doubled process accounting times
- Minor feature updates to the BCM1250/BCM1480 header files
- bcm1x80: while 4 CPUs are found, Bogomips are never displayed. So
maybe something with the recent SMP change is still not 100% right.
[Actually, it seems this is not reported in non-SMP either]
- bcm1x80: PCI DMA seems completely broken.
- loading an initrd works (needs CoLo 1.21)
- PCI works: SCSI, USB, audio and network tested
- loading of initrd image generated by tip22 works
- HAL2 (sound) works, but driver should be converted to ALSA
- VINO: supposedly works with a 32-bit kernel
- fb: unknown
- loading of initrd image generated by tip32 works
- fb works (with 4 MB)
Broadcom SWARM (1250):
- loading an initrd works
- IDE works