Hello everyone. I am writing a device driver on qemu using a PCI card that I
modelled bare functionality in so we can test the driver while the board is
made. I have a couple of questions that I can't find having been addressed in
the past.
The system is x86_64 running Ubuntu. I am using KVM on my gentoo x86_64 machine.
It seems that a dma transfer is instant in qemu as far as reality is concerned.
The only way to make the guest see some time passing before some kind of dma
interrupt complete fires is to wait to fire the interrupt. I did this using a
timer. It seems I should use vm_clock for my timer, but this clock increments
even when I should not expect it to. For example, in my MemoryRegion's write
ops, if I do printf("time is %lld\n", qemu_get_clock_ns(vm_clock)); twice, I
will see a change in the clock. If I put a sleep(1) here the whole simulation
will freeze, but time continues.
So when I use a timer based on vm_clock, and I want to have it expire every 20
us (arbitrary selected time) to start the next page in my transfer (the modeled
card does scatter/gather DMA) I wind up where 2500 timers that should take 50ms
of simulation time actually take many seconds of simulation time because the
timer will fire anywhere from ~20-300 microseconds simulation time.
This does not happen if I am not using KVM. It instead seems to increment
outside this context. However, this becomes too slow to be usable. (Maybe there
are other options for proceeding without KVM that I am not aware of?)
Is there a way I can achieve timing that is synced to actual CPU clock ticks in
my devices? I have looked through the code but not found anything.
Here are my start parameters for qemu:
./qemu_bld/x86_64-softmmu/qemu-system-x86_64 -m 1024 -drive
file=/home/path/ubuntu.img,if=virtio -net nic,model=virtio -net user -vga std
-device pci_test -fsdev local,id=work,path=/home/path/,security_model=mapped
-device virtio-9p-pci,fsdev=work,mount_tag=work --enable-kvm
Thanks all.
Steven
