Windows: DeviceTree traversal/offset bugfixes #1683
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
This updates the `devicetree` plugin to return the virtual offsets of the device objects themselves instead of that of their parent driver.
Previously, the attached devices were being traversed once depth-wise, but the horizontal levels of the device tree were not being properly traversed, leading to tons of missing device entries. This fixes the issue by implementing a recursive staticmethod for properly descending + enumerating devices on each level of the tree.
|
@ikelos One consideration for this should be what to do about the |
dgmcdona
force-pushed
the
devicetree_traversal_and_offsets
branch
from
ebf2d6f to
a1091c8
Compare
dgmcdona
force-pushed
the
devicetree_traversal_and_offsets
branch
3 times, most recently
from
1b024c0 to
dea091a
Compare
Moves a lot of the code that was parsing information about device objects out of the `DeviceTree` plugin logic and into the `DEVICE_OBJECT` extension class. Also moved the dict mapping integer values to device types into the extension module. Fixes a bug within the extension class' `get_attached_devices` method - now, immediate children of the device in the tree are yielded, instead of the faulty deep traversal that was occuring before. Because of this bugfix in an extension class as well as the addition of new methods in the extension class, this also bumps the framework minor version number.
dgmcdona
force-pushed
the
devicetree_traversal_and_offsets
branch
from
dea091a to
73e422f
Compare
|
I'd pick a different method name, and run them alongside each other (deprecating the old one)? Eventually, if it's really worth it, you can shift the new one back to the old name. If the code belongs centrally, it's much better to have it live there than in a plugin where someone may not know about it and try to reimplement (possibly incorrectly). If it's only for the generator (ie, not to be used by other plugins) then it's fine. You could also consider a visitor design pattern, that takes a function and runs it on each node in the tree in a particular order, calling it with the node and the depth at each point? |
|
Thanks, marking as draft again while I revisit this. |
There are two issues with the current implementation of the
windows.devicetree.DeviceTreeplugin:DEVICE_OBJECT.get_attached_devicesextension method does not traverse the full breadth at each level of the device tree through theNextDevicelinked-list.This PR fixes the issue by implementing a recursive, private staticmethod on the
DeviceTreeplugin class that does a complete depth-first traversal of the tree, yielding many more device objects (using their virtual address as theOffsetmember instead of that of the parent driver).