Hi Drew, On Tue, Feb 15, 2022 at 04:50:37PM +0100, Andrew Jones wrote: > On Tue, Feb 15, 2022 at 02:16:32PM +0000, Alexandru Elisei wrote: > > Hi Drew, > > > > On Tue, Feb 15, 2022 at 01:53:00PM +0100, Andrew Jones wrote: > > > On Tue, Feb 15, 2022 at 10:07:16AM +0000, Alexandru Elisei wrote: > > > > > > > > I've started working on the next iteration of the kvmtool test > > > > runner support series, I'll do my best to make sure kvmtool wll be able to run > > > > the tests when kvm-unit-tests has been configured with --arch=arm. > > > > > > > > > > Excellent! > > > > > > BTW, I went ahead an pushed a patch to misc/queue to improve the initrd > > > address stuff > > > > > > https://gitlab.com/rhdrjones/kvm-unit-tests/-/commit/6f8f74ed2d9953830a3c74669f25439d9ad68dec > > > > > > It may be necessary for you if kvmtool shares its fdt creation between > > > aarch64 and aarch32 guests, emitting 8 byte initrd addresses for both, > > > even though the aarch32 guest puts the fdt below 4G. > > > > While trying your patch (it works, but with the caveat below), I remembered that > > kvmtool is not able to run kvm-unit-tests for arm. That's because the code is > > not relocatable (like it is for arm64) and the text address is hardcoded in the > > makefile. > > > > In past, to run the arm tests with kvmtool, I was doing this change: > > > > diff --git a/arm/Makefile.arm b/arm/Makefile.arm > > index 3a4cc6b26234..6c580b067413 100644 > > --- a/arm/Makefile.arm > > +++ b/arm/Makefile.arm > > @@ -14,7 +14,7 @@ CFLAGS += $(machine) > > CFLAGS += -mcpu=$(PROCESSOR) > > CFLAGS += -mno-unaligned-access > > > > -arch_LDFLAGS = -Ttext=40010000 > > +arch_LDFLAGS = -Ttext=80008000 > > > > define arch_elf_check = > > endef > > > > Any suggestions how to fix that? One way would be to change the LDFLAGS based on > > $TARGET. Another way would be to make the arm tests relocatable, I tried to do > > that in the past but I couldn't make any progress. > > Ideally we'd eventually make it relocatable, but it's not worth moving > mountains, so I'm OK with the LDFLAGS approach. > > > > > Separate from that, I also tried to run the 32 bit arm tests with run_tests.sh. > > The runner uses qemu-system-arm (because $ARCH_NAME is arm in > > scripts/arch-run.bash::search_qemu_binary()), but uses kvm as the accelerator > > (because /dev/kvm exists in scrips/arch-run.bash::kvm_available()). This fails > > with the error: > > > > qemu-system-arm: -accel kvm: invalid accelerator kvm > > > > I don't think that's supposed to happen, as kvm_available() specifically returns > > true if $HOST = aarch64 and $ARCH = arm. Any suggestions? > > > > Ah, that's a kvm-unit-tests bug. Typically qemu-system-$ARCH_NAME is > correct and, with TCG, qemu-system-arm runs the arm built unit tests fine. > But, when KVM is enabled, the QEMU used should be qemu-system-$HOST. > > I don't think we want to change search_qemu_binary(), though, as that > would require ACCEL being set first and may not apply to all > architectures. Probably the best thing to do is fix this up in arm/run Definitely, this should go in arm/run. > > diff --git a/arm/run b/arm/run > index 2153bd320751..ff18def43403 100755 > --- a/arm/run > +++ b/arm/run > @@ -13,6 +13,10 @@ processor="$PROCESSOR" > ACCEL=$(get_qemu_accelerator) || > exit $? > > +if [ "$ACCEL" = "kvm" ] && [ -z "$QEMU" ]; then > + QEMU=qemu-system-$HOST > +fi Looking at get_qemu_accelerator, I think it returns true when /dev/kvm exists, which means that get_qemu_accelerator can return "kvm" when trying to run the arm64 tests on an arm machine with KVM. But in this case, it's impossible to run the tests using KVM. And I also think there should be some checking that aarch64=off works, as there are arm64 CPUs which don't support AArch32 (like Cortex-A510). Thanks, Alex > + > qemu=$(search_qemu_binary) || > exit $? > > > Thanks, > drew >
