Consolidate worker threads #317
Conversation
jakecorrenti
requested review from
MatiasVara,
mtjhrc,
slp and
tylerfanelli
as code owners
|
The SEV-SNP update also has a worker thread. Can we merge that first and rebase on top? |
jakecorrenti
force-pushed
the
worker-thread
branch
from
701042f to
d0c4c40
Compare
Sure. I figured I'd at least post the changes and let people review if they want |
|
Sure, that works. Can you mark as draft to note that its waiting on another PR? |
|
I think this mechanism can also be used for arm-cca. There is also a worker although the mechanism has not been generalized as here. |
jakecorrenti
force-pushed
the
worker-thread
branch
from
d0c4c40 to
f8c2a7f
Compare
jakecorrenti
force-pushed
the
worker-thread
branch
6 times, most recently
from
ab523d2 to
edcf800
Compare
|
Disclaimer: I can't test these changes on a SEV machine to verify the TEE code works |
|
Tested on SEV-SNP. LGTM! Can you rebase on |
Rather than spawning a new worker thread every time some functionality needs it, which results in it's own Message type, create something generic that can be used in one thread. Signed-off-by: Jake Correnti <[email protected]>
Rather than creating the worker thread on macOS differently than we do
for the other x86 tasks, use the same associative function by taking
advnatage of the generic `WorkerMessage` enum.
Signed-off-by: Jake Correnti <[email protected]>
On x86, use the same sender as we would for macOS. Additionally, rather than using an EventFd to determine when the thread work is done, use a response sender/receiver like macOS. Signed-off-by: Jake Correnti <[email protected]>
For TEE VMs, use the same sender as we would for macOS or an x86 VM with a split IRQCHIP. ADditionally, use a channel for inter-process communication instead of an EventFd. Signed-off-by: Jake Correnti <[email protected]>
jakecorrenti
force-pushed
the
worker-thread
branch
from
edcf800 to
ddde951
Compare
Done! |
|
@MatiasVara PTAL. I made some modifications to the worker thread for TEEs |
Whenever the vCPU needs to access resources the Vm or Vmm has control over, we end up spawning a worker thread to complete the task. Rather than spawning a new worker thread for each task, use a single thread, but make the message that's passed generic to allow for multiple use cases.