Postpone the setting of the /memory node so that any allocations made
in course of the devicetree prep (e.g. the SIO reservations allocated
from the top of memory) are reflected in the usable memory span that's
set on the /memory node.
Signed-off-by: Martin Povišer <povik@cutebit.org>
On kboot, parse the ADT for SIO firmware parameters, set up the expected
data structures for SIO consumption in reserved memory, and pass those
in properties of the SIO FDT node.
Signed-off-by: Martin Povišer <povik@cutebit.org>
Make dt_device_set_reserved_mem take in an explicit iova instead of
reading it off from a DART. Add a separate 'from_dart' variety of the
function for getting the iova from DART.
Signed-off-by: Martin Povišer <povik@cutebit.org>
The new function parses ADT's segment-ranges and builds up nodes in
/reserved-memory representing the appropriate physical and device
address spans. This is to describe the memory placement of ASC firmware.
Signed-off-by: Martin Povišer <povik@cutebit.org>
This doesn't make any sense... M2 Ultra should be 2x M2 Max, so there's
no reason for this constant to be different.
I get the feeling Apple messed this one up too...
Also clean up some other constants.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Add a warning message when overriding the target serial device with
SECDEV variable, just in case one is caught by surprise there.
Signed-off-by: Alexis Deruelle <alexis.deruelle@gmail.com>
Under macOS host, secondary UART device name is different from the linux
device name. Check OS name and chose default device name accordingly.
Signed-off-by: Alexis Deruelle <alexis.deruelle@gmail.com>
- in 'main' section, device name and baud values are redundant as the
UartInterface constructor handles it already
- Add a comment on how to run proxy.py as script
Signed-off-by: Alexis Deruelle <alexis.deruelle@gmail.com>
m1n1 UART device names on macOS follow /dev/cu.usbmodemP_XX naming
convention. Check for platform name using platform.system() and use the
macOS device name instead of /dev/m1n1 if equals 'Darwin'
Signed-off-by: Alexis Deruelle <alexis.deruelle@gmail.com>
This is spammy and sometimes so slow it blocks forever. Let's explicitly
remove these ranges from the GPUVM tracer.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Since we directly trace the queue rptr/wptr memory locations, we don't
need the ASC tracing and this speeds things up quite a bit (without it,
Linux can get stuck in a pathological stats gathering loop without
making forward progress while traced).
Signed-off-by: Asahi Lina <lina@asahilina.net>
