Re: [libvirt] [PATCH qom-cpu 01/11] target-i386: Don't set any KVM flag by default if KVM is disabled

On Mon, Jan 07, 2013 at 02:15:59PM +0200, Gleb Natapov wrote: > On Mon, Jan 07, 2013 at 10:09:24AM -0200, Eduardo Habkost wrote: > > On Mon, Jan 07, 2013 at 01:42:53PM +0200, Gleb Natapov wrote: > > > On Mon, Jan 07, 2013 at 09:42:36AM -0200, Eduardo Habkost wrote: > > > > On Sun, Jan 06, 2013 at 01:32:34PM +0200, Gleb Natapov wrote: > > > > > On Fri, Jan 04, 2013 at 08:01:02PM -0200, Eduardo Habkost wrote: > > > > > > This is a cleanup that tries to solve two small issues: > > > > > > > > > > > > - We don't need a separate kvm_pv_eoi_features variable just to keep a > > > > > > constant calculated at compile-time, and this style would require > > > > > > adding a separate variable (that's declared twice because of the > > > > > > CONFIG_KVM ifdef) for each feature that's going to be enabled/disable > > > > > > by machine-type compat code. > > > > > > - The pc-1.3 code is setting the kvm_pv_eoi flag on cpuid_kvm_features > > > > > > even when KVM is disabled at runtime. This small incosistency in > > > > > > the cpuid_kvm_features field isn't a problem today because > > > > > > cpuid_kvm_features is ignored by the TCG code, but it may cause > > > > > > unexpected problems later when refactoring the CPUID handling code. > > > > > > > > > > > > This patch eliminates the kvm_pv_eoi_features variable and simply uses > > > > > > CONFIG_KVM and kvm_enabled() inside the enable_kvm_pv_eoi() compat > > > > > > function, so it enables kvm_pv_eoi only if KVM is enabled. I believe > > > > > > this makes the behavior of enable_kvm_pv_eoi() clearer and easier to > > > > > > understand. > > > > > > > > > > > > Signed-off-by: Eduardo Habkost <ehabkost(a)redhat.com&gt; > > > > > > --- > > > > > > Cc: kvm(a)vger.kernel.org > > > > > > Cc: Michael S. Tsirkin <mst(a)redhat.com&gt; > > > > > > Cc: Gleb Natapov <gleb(a)redhat.com&gt; > > > > > > Cc: Marcelo Tosatti <mtosatti(a)redhat.com&gt; > > > > > > > > > > > > Changes v2: > > > > > > - Coding style fix > > > > > > --- > > > > > > target-i386/cpu.c | 8 +++++--- > > > > > > 1 file changed, 5 insertions(+), 3 deletions(-) > > > > > > > > > > > > diff --git a/target-i386/cpu.c b/target-i386/cpu.c > > > > > > index 82685dc..e6435da 100644 > > > > > > --- a/target-i386/cpu.c > > > > > > +++ b/target-i386/cpu.c > > > > > > @@ -145,15 +145,17 @@ static uint32_t kvm_default_features = (1 << KVM_FEATURE_CLOCKSOURCE) | > > > > > > (1 << KVM_FEATURE_ASYNC_PF) | > > > > > > (1 << KVM_FEATURE_STEAL_TIME) | > > > > > > (1 << KVM_FEATURE_CLOCKSOURCE_STABLE_BIT); > > > > > > -static const uint32_t kvm_pv_eoi_features = (0x1 << KVM_FEATURE_PV_EOI); > > > > > > #else > > > > > > static uint32_t kvm_default_features = 0; > > > > > > -static const uint32_t kvm_pv_eoi_features = 0; > > > > > > #endif > > > > > > > > > > > > void enable_kvm_pv_eoi(void) > > > > > > { > > > > > > - kvm_default_features |= kvm_pv_eoi_features; > > > > > > +#ifdef CONFIG_KVM > > > > > You do not need ifdef here. > > > > > > > > We need it because KVM_FEATURE_PV_EOI is available only if CONFIG_KVM is > > > > set. > > > > > > > > I could also write it as: > > > > > > > > if (kvm_enabled()) { > > > > #ifdef CONFIG_KVM > > > > kvm_default_features |= (1UL << KVM_FEATURE_PV_EOI); > > > > #endif > > > > } > > > > > > > > But I find it less readable. > > > > > > > > > > > Why not define KVM_FEATURE_PV_EOI unconditionally? > > > > It comes from the KVM kernel headers, that are included only if > > CONFIG_KVM is set, and probably won't even compile in non-Linux systems. > > > > I have a dejavu feeling. I believe we had this exact problem before, > > maybe about some other #defines that come from the Linux KVM headers and > > won't be available in non-Linux systems. > > > It is better to hide all KVM related differences somewhere in the > headers where no one sees them instead of sprinkle them all over the > code. We can put those defines in include/sysemu/kvm.h in !CONFIG_KVM > part. Or have one ifdef CONFIG_KVM at the beginning of the file and > define enable_kvm_pv_eoi() there and provide empty stub otherwise. If we had an empty enable_kvm_pv_eoi() stub, we would need an #ifdef around the real implementation. I mean, I don't think this: #ifdef CONFIG_KVM int enable_kvm_pv_eoi() { [...] } #endif

is any better than this: int enable_kvm_pv_eoi() { #ifdef CONFIG_KVM [...] #endif } So this is probably a good reason to duplicate the KVM_FEATURE_* #defines in the QEMU code, instead?