[libvirt] [PATCH 4/4] qemu: command: wire up usage of q35/ich9 disable s3/s4

[Cole Robinson]

On 01/10/2016 04:54 AM, Martin Kletzander wrote:
> On Sat, Jan 09, 2016 at 04:32:23PM -0500, Cole Robinson wrote:
>> If the q35 specific disable s3/s4 setting isn't disabled, fallback to
> 
> this reads weirdly, maybe you meant "supported" instead of "disabled"?
> 

Yeah, I meant supported. Fixed

>> specifying the PIIX setting, which is the previous behavior. It doesn't
>> have any effect, but qemu will just warn about it rather than error:
>>
>> qemu-system-x86_64: Warning: global PIIX4_PM.disable_s3=1 not used
>> qemu-system-x86_64: Warning: global PIIX4_PM.disable_s4=1 not used
>>
>> Since it doesn't error, I don't think we should either, since there
>> may be configs in the wild that already have q35 + disable_s3/4 (via
>> virt-manager)
> 
> We can even skip specifying it at all, back when I implemented it there
> were just no nono-pc x86 machines types =) Or we can error out when
> starting as we do with other changes that would otherwise be
> incompatible.  If the error message is clear, I see no problem with it.
> But that can be changed later on.
> 

Yeah I'm still a little scared of rejecting configs that plausibly exist in
the wild, when the end result is the same either (PM wasn't disabled). We can
add it later though

>> ---
>> src/qemu/qemu_command.c                            | 32 ++++++++++++++++++----
>> .../qemuxml2argv-q35-pm-disable-fallback.args      | 23 ++++++++++++++++
>> .../qemuxml2argv-q35-pm-disable-fallback.xml       | 18 ++++++++++++
>> .../qemuxml2argv-q35-pm-disable.args               | 23 ++++++++++++++++
>> .../qemuxml2argv-q35-pm-disable.xml                | 18 ++++++++++++
>> tests/qemuxml2argvtest.c                           |  9 ++++++
>> 6 files changed, 117 insertions(+), 6 deletions(-)
>> create mode 100644
>> tests/qemuxml2argvdata/qemuxml2argv-q35-pm-disable-fallback.args
>> create mode 100644
>> tests/qemuxml2argvdata/qemuxml2argv-q35-pm-disable-fallback.xml
>> create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-pm-disable.args
>> create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-q35-pm-disable.xml
>>
>> diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c
>> index 0ee3247..dc7adcb 100644
>> --- a/src/qemu/qemu_command.c
>> +++ b/src/qemu/qemu_command.c
>> @@ -9806,25 +9806,45 @@ qemuBuildCommandLine(virConnectPtr conn,
>>     }
>>
>>     if (def->pm.s3) {
>> -        if (!virQEMUCapsGet(qemuCaps, QEMU_CAPS_PIIX_DISABLE_S3)) {
>> +        const char *pm_object = NULL;
>> +
>> +        if (qemuDomainMachineIsQ35(def) &&
>> +            virQEMUCapsGet(qemuCaps, QEMU_CAPS_ICH9_DISABLE_S3)) {
>> +            pm_object = "ICH9-LPC";
>> +        } else if (virQEMUCapsGet(qemuCaps, QEMU_CAPS_PIIX_DISABLE_S3)) {
>> +            /* We fall back to this even for q35, since it's what we did
>> +               pre-q35-pm support. QEMU starts up fine (with a warning) if
>> +               mixing PIIX PM and -M q35. Starting to reject things here
>> +               could mean we refuse to start existing configs in the wild.*/
>> +            pm_object = "PIIX4_PM";
>> +        } else {
>>             virReportError(VIR_ERR_CONFIG_UNSUPPORTED,
>>                            "%s", _("setting ACPI S3 not supported"));
>>             goto error;
>>         }
>> +
>>         virCommandAddArg(cmd, "-global");
>> -        virCommandAddArgFormat(cmd, "PIIX4_PM.disable_s3=%d",
>> -                               def->pm.s3 == VIR_TRISTATE_BOOL_NO);
>> +        virCommandAddArgFormat(cmd, "%s.disable_s3=%d",
>> +                               pm_object, def->pm.s3 == VIR_TRISTATE_BOOL_NO);
>>     }
>>
>>     if (def->pm.s4) {
>> -        if (!virQEMUCapsGet(qemuCaps, QEMU_CAPS_PIIX_DISABLE_S4)) {
>> +        const char *pm_object;
>> +
> 
> In the previous block you initialize it, but here you don't.  How about
> putting it out of these blocks and just initialize it to:
> 
> pm_object = qemuDomainMachineIsQ35(def) ? "ICH9-LPC" : "PIIX4_PM";

That doesn't work as is with my patch since it doesn't incorporate checking
qemuCaps. I made a similarish change though and pushed. Thanks for the review!

- Cole