> On Apr 10, 2020, at 3:20 AM, Liran Alon <liran.alon@xxxxxxxxxx> wrote:
>
>
> On 10/04/2020 6:52, Nadav Amit wrote:
>> 2. Try to run the tests with more than 4GB of memory. The last time I tried
>> (actually by running the test on bare metal), the INIT test that Liran
>> wrote failed.
> Wasn't this test failure fixed with kvm-unit-test commit fc47ccc19612 ("x86: vmx: Verify pending LAPIC INIT event consume when exit on VMX_INIT")?
> If not, can you provide the details of this new failure? As I thought this commit address the previous issue you have reported when running this test
> on bare-metal.
Your patch solved the problem of INIT and apparently you got the right
implementation in KVM.
There appears to be another issue, which I suspect is only a test issue,
when I run the tests on bare-metal with more than 4GB of memory. If I remove
Paolo’s patch the enabled support for more than 4GB of RAM, or if I run it
on a VM with 4GB of RAM it passes. I did not run the tests on KVM - to be
fair.
Here is the splat I got on a non-KVM hypervisor with 8GB or RAM:
Test suite: vmx_init_signal_test
Unhandled cpu exception 14 #PF at ip 0000000000419698
PF at 0x419698 addr 0x102066000
error_code=0000 rflags=00010046 cs=00000008
rax=0000000000000000 rcx=000000000000080b rdx=0000000000000001 rbx=0000000102066000
rbp=000000000053a7b8 rsi=0000000000000000 rdi=0000000102065000
r8=000000000053a016 r9=00000000000003f8 r10=000000000000000d r11=0000000000000000
r12=0000000102065000 r13=0000000102065000 r14=0000000000000000 r15=0000000000000000
cr0=0000000080000011 cr2=0000000102066000 cr3=0000000000422000 cr4=0000000000000020
cr8=0000000000000000
STACK: @419698 407539 419f81 419eb1
[ I ran it also on bare metal, and got IIRC got a similar splat, but did not
save it. ]
The splat shows the failure happened in alloc_page() which was called from
init_signal_test_thread() on the following line:
freelist = *(void **)freelist;
( and IIRC freelist is nil )
I did not have time to analyze the problem or even run it on KVM.
