- Add IOBusDevice (nvram_addr_hi_dev) for NVRAM addr hi.
- Add IOBusDevice (nvram_dev) for NVRAM data.
- Make all IOBusDevices use the same code.
- Log error if 4 least significant bits of offset are not zero.
- Correctly byte swap the value before passing it to the IOBusDevice.
- When reading, duplicate the bytes in a word or dword like a real Power Mac does.
- Add all the interrupts including DMA.
- Modify the Interrupt to IRQ_ID translation so the interrupts belonging to the first set of 32 interrupts don't need to be shifted.
Otherwise if pull_data is called again, it will think that it still
has data available in the buffer (rem_len will be non-zero) and
random data at the buffer location will be returned.
This manifested itself as noise being played back in the JS
implementation of the SoundServer. The cubeb implementation was not
affected because it stops polling once it's told it has no more
data in the buffer. Both approaches are valid (the JS version pads
data with silence), and the DMA buffer should support both.
Result of running IWYU (https://include-what-you-use.org/) and
applying most of the suggestions about unncessary includes and
forward declarations.
Was motivated by observing that <thread> was being included in
ppcopcodes.cpp even though it was unused (found while researching
the use of threads), but seems generally good to help with build
times and correctness.
For some reason, on Windows, whether it be VS2022 or Clang, it will crash when trying to write to a Big Mac register if you don't specify the exact name of the component within Heathrow.
This commit fixes that.
