- PS/2 answers with ERROR if PS/2 mouse is not connected, read_ack should handle error cases correctly. We should also probe correctly for the mouse. (0xFE on all commands) - bochs shows us distorted text console output, why? - bootloader has trouble detecting drive geometry of hard disks, only the floppy mode works reliably at the moment (in emulators and on real machines). Especially USB-pens in floppy BIOS emulation seems to report wrong CHS values. - GUI: sort out a readable and type-safe (as much as possible) modus operandi. Accesses via base member chains are type safe, but we have to know the depth of the hierarchy of a derived class. Casts on the other hand are shorter to read, but not type-safe. - real hardware like the Asus EEEPC 701 cannot switch A20 correctly, no clue why? - if a task ends, what should we do? currently we crash - memorymanagement should be a library of the libc or even a standalone library, not part of the kernel directory, get rid of kernel_panic there (this requires the kernel_stub.c kernel_panic to exist in libc tests) - clang and -O2/-O3 break test_printf.c and printf, va_arg macros break. inlining va_arg functions should be possible. -Wno-inline, attribute(used), attribute(noinline) have no effect (or not always, depends on the host platform?). - should we unify the init interface of all drivers? Currently only the PCI ones are unified (for instance by having the event handler as a register_handler method and not part of the ctor). Arguments for splitting: - also the PS/2 devices are on a serio bus - there are ISA/AGP/I2C/USB buses which are also dynamic - we could have multiple registered handlers for some devices - rtl8139 driver: - can only transmit one packet in parallel currently (transmitting boolean flag. Taken from Tyndur. No queueing currently and postponing of sending of packets) - networking: - strict layering needed, we should really not see the MAC in IPv4 and UDP, only in ARP