Since I knew memory often is more painful - start with 512MB, 1CPU, 1 PCI Passthrough
Note: I installed debug symbols for glibc and qemu
On init I find initially the guests CPU thread rather busy (well, booting up)
80.66% CPU 0/KVM [kernel]
Passthrough is successful - lspci from guest:
00:07.0 3D controller: NVIDIA Corporation Device 1db8 (rev a1)
Usually you have a few sections of execution, here are my timings:
1. from virsh start until there is a console connected
That is the libvirt prep and the early startup of the qemu process
until it provides something on the socket representing the console
=> 4 seconds
2. from #1 until we see anything in the console moving
That usually is mostly consumed by KVM (host) to prep mappings
and potentially scrub the memory and prepare devices.
=> 12 seconds
3. from #2 until we can log in
That is the guest kernel, followed by guest userspace initializing
=> 17 seconds
So I overall needed 33 seconds for a guest with one device passed through, that is not too bad.
Your example was the full size (1.4TB) guest and I'd expect that this is what happens here.
Trying a few more tests ...
Since I knew memory often is more painful - start with 512MB, 1CPU, 1 PCI Passthrough
Note: I installed debug symbols for glibc and qemu
On init I find initially the guests CPU thread rather busy (well, booting up)
80.66% CPU 0/KVM [kernel]
Passthrough is successful - lspci from guest:
00:07.0 3D controller: NVIDIA Corporation Device 1db8 (rev a1)
Usually you have a few sections of execution, here are my timings:
1. from virsh start until there is a console connected
That is the libvirt prep and the early startup of the qemu process
until it provides something on the socket representing the console
=> 4 seconds
2. from #1 until we see anything in the console moving
That usually is mostly consumed by KVM (host) to prep mappings
and potentially scrub the memory and prepare devices.
=> 12 seconds
3. from #2 until we can log in
That is the guest kernel, followed by guest userspace initializing
=> 17 seconds
So I overall needed 33 seconds for a guest with one device passed through, that is not too bad.
Your example was the full size (1.4TB) guest and I'd expect that this is what happens here.
Trying a few more tests ...