Re: [RFC PATCH v4 08/15] tests/qtest: Adjust test-hmp to always pass -cpu option

[Fabiano Rosas <farosas@suse.de>]

"Dr. David Alan Gilbert" <dgilbert@redhat.com> writes:

> * Fabiano Rosas (farosas@suse.de) wrote:
>> Signed-off-by: Fabiano Rosas <farosas@suse.de>
>> ---
>>  tests/qtest/test-hmp.c | 34 +++++++++++++++++++++++++++++++---
>>  1 file changed, 31 insertions(+), 3 deletions(-)
>> 
>> diff --git a/tests/qtest/test-hmp.c b/tests/qtest/test-hmp.c
>> index f8b22abe4c..c367612d4a 100644
>> --- a/tests/qtest/test-hmp.c
>> +++ b/tests/qtest/test-hmp.c
>> @@ -121,21 +121,49 @@ static void test_info_commands(QTestState *qts)
>>      g_free(info_buf);
>>  }
>>  
>> +static const char *arch_get_cpu(const char *machine)
>> +{
>> +    const char *arch = qtest_get_arch();
>> +
>> +    if (g_str_equal(arch, "aarch64")) {
>> +        if (!strncmp(machine, "virt", 4)) {
>> +            return "cortex-a57";
>
> Won't that break on a kvm host on a different CPU?
> Would -cpu max   work on everything?

These tests run with -accel qtest only. No TCG or KVM. So this won't
break a KVM host with a different CPU. But we also cannot expect that
cpu to do anything useful. This is more to avoid initial configuration
code to break due to lack of cpu.

About -cpu max, before this series, it works because of a nuance of the
code which does:

aarch64_max_initfn:
if kvm || hvf
    host
else
    cortex-a57
    extra_features that depend on TCG

After this series this becomes roughly (see patch 01):

aarch64_max_initfn:
if kvm || hvf
    host
else if qtest
    cortex-a57
#ifdef CONFIG_TCG    
else if tcg
    cortex-a57
    extra_features that depend on TCG
#endif

The above routine causes us to have two different -cpu max depending on
whether TCG is built in or not, so I'm basically bypassing it and
hardcoding cortex-a57 to be consistent.