mirror of
https://github.com/torvalds/linux.git
synced 2024-12-27 13:22:23 +00:00
e4ac58afdf
Saves like 1,600 lines of code, is way easier to debug, compilers frequently do a better job than the cut and paste type of handlers many boards had. And finally having all the stuff done in a single place also means alot of bug potencial for the MT ASE is gone. The only surviving handler in assembler is the DECstation one; I hope Maciej will rewrite it. Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
20 lines
1.0 KiB
Plaintext
20 lines
1.0 KiB
Plaintext
1. Need to figure out why PCI writes to the IOC3 hang, and if it is okay
|
|
not to write to the IOC3 ever.
|
|
2. Need to figure out RRB allocation in bridge_startup().
|
|
3. Need to figure out why address swaizzling is needed in inw/outw for
|
|
Qlogic scsi controllers.
|
|
4. Need to integrate ip27-klconfig.c:find_lboard and
|
|
ip27-init.c:find_lbaord_real. DONE
|
|
5. Is it okay to set calias space on all nodes as 0, instead of 8k as
|
|
in irix?
|
|
6. Investigate why things do not work without the setup_test() call
|
|
being invoked on all nodes in ip27-memory.c.
|
|
8. Too many do_page_faults invoked - investigate.
|
|
9. start_thread must turn off UX64 ... and define tlb_refill_debug.
|
|
10. Need a bad pmd table, bad pte table. __bad_pmd_table/__bad_pagetable
|
|
does not agree with pgd_bad/pmd_bad.
|
|
11. All intrs (ip27_do_irq handlers) are targetted at cpu A on the node.
|
|
This might need to change later. Only the timer intr is set up to be
|
|
received on both Cpu A and B. (ip27_do_irq()/bridge_startup())
|
|
13. Cache flushing (specially the SMP version) has to be investigated.
|