Re: [PATCH v1 RFC 6/6] KVM: s390: add cpu model support

From: Alexander Graf
Date: Mon May 19 2014 - 16:14:11 EST



On 19.05.14 19:03, Michael Mueller wrote:
On Mon, 19 May 2014 16:49:28 +0200
Alexander Graf <agraf@xxxxxxx> wrote:

On 19.05.14 16:18, Michael Mueller wrote:
On Mon, 19 May 2014 13:48:08 +0200
Alexander Graf <agraf@xxxxxxx> wrote:

On 19.05.14 12:53, Michael Mueller wrote:
On Fri, 16 May 2014 22:31:12 +0200
Alexander Graf <agraf@xxxxxxx> wrote:

On 16.05.14 17:39, Michael Mueller wrote:
On Fri, 16 May 2014 14:08:24 +0200
Alexander Graf <agraf@xxxxxxx> wrote:

On 13.05.14 16:58, Michael Mueller wrote:
This patch enables cpu model support in kvm/s390 via the vm attribute
interface.

During KVM initialization, the host properties cpuid, IBC value and the
facility list are stored in the architecture specific cpu model structure.

During vcpu setup, these properties are taken to initialize the related SIE
state. This mechanism allows to adjust the properties from user space and thus
to implement different selectable cpu models.

This patch uses the IBC functionality to block instructions that have not
been implemented at the requested CPU type and GA level compared to the
full host capability.

Userspace has to initialize the cpu model before vcpu creation. A cpu model
change of running vcpus is currently not possible.
Why is this VM global? It usually fits a lot better modeling wise when
CPU types are vcpu properties.
It simplifies the code substantially because it inherently guarantees the vcpus being
configured identical. In addition, there is no S390 hardware implementation containing
inhomogeneous processor types. Thus I consider the properties as machine specific.

Signed-off-by: Michael Mueller <mimu@xxxxxxxxxxxxxxxxxx>
---
arch/s390/include/asm/kvm_host.h | 4 +-
arch/s390/include/uapi/asm/kvm.h | 23 ++++++
arch/s390/kvm/kvm-s390.c | 146 ++++++++++++++++++++++++++++++++++++++-
arch/s390/kvm/kvm-s390.h | 1 +
4 files changed, 172 insertions(+), 2 deletions(-)

diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index b4751ba..6b826cb 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -84,7 +84,8 @@ struct kvm_s390_sie_block {
atomic_t cpuflags; /* 0x0000 */
__u32 : 1; /* 0x0004 */
__u32 prefix : 18;
- __u32 : 13;
+ __u32 : 1;
+ __u32 ibc : 12;
__u8 reserved08[4]; /* 0x0008 */
#define PROG_IN_SIE (1<<0)
__u32 prog0c; /* 0x000c */
@@ -418,6 +419,7 @@ struct kvm_s390_cpu_model {
unsigned long *sie_fac;
struct cpuid cpu_id;
unsigned long *fac_list;
+ unsigned short ibc;
};
struct kvm_arch{
diff --git a/arch/s390/include/uapi/asm/kvm.h b/arch/s390/include/uapi/asm/kvm.h
index 313100a..82ef1b5 100644
--- a/arch/s390/include/uapi/asm/kvm.h
+++ b/arch/s390/include/uapi/asm/kvm.h
@@ -58,12 +58,35 @@ struct kvm_s390_io_adapter_req {
/* kvm attr_group on vm fd */
#define KVM_S390_VM_MEM_CTRL 0
+#define KVM_S390_VM_CPU_MODEL 1
/* kvm attributes for mem_ctrl */
#define KVM_S390_VM_MEM_ENABLE_CMMA 0
#define KVM_S390_VM_MEM_CLR_CMMA 1
#define KVM_S390_VM_MEM_CLR_PAGES 2
+/* kvm attributes for cpu_model */
+
+/* the s390 processor related attributes are r/w */
+#define KVM_S390_VM_CPU_PROCESSOR 0
+struct kvm_s390_vm_cpu_processor {
+ __u64 cpuid;
+ __u16 ibc;
+ __u8 pad[6];
+ __u64 fac_list[256];
+};
+
+/* the machine related attributes are read only */
+#define KVM_S390_VM_CPU_MACHINE 1
+struct kvm_s390_vm_cpu_machine {
+ __u64 cpuid;
+ __u32 ibc_range;
+ __u8 pad[4];
+ __u64 fac_mask[256];
+ __u64 hard_fac_list[256];
+ __u64 soft_fac_list[256];
+};
+
/* for KVM_GET_REGS and KVM_SET_REGS */
struct kvm_regs {
/* general purpose regs for s390 */
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index a53652f..9965d8b 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -369,6 +369,110 @@ static int kvm_s390_mem_control(struct kvm *kvm, struct
kvm_device_attr *attr) return ret;
}
+static int kvm_s390_set_processor(struct kvm *kvm, struct kvm_device_attr *attr)
+{
+ struct kvm_s390_vm_cpu_processor *proc;
+
+ if (atomic_read(&kvm->online_vcpus))
+ return -EBUSY;
+
+ proc = kzalloc(sizeof(*proc), GFP_KERNEL);
+ if (!proc)
+ return -ENOMEM;
+
+ if (copy_from_user(proc, (void __user *)attr->addr,
+ sizeof(*proc))) {
+ kfree(proc);
+ return -EFAULT;
+ }
+
+ mutex_lock(&kvm->lock);
+ memcpy(&kvm->arch.model.cpu_id, &proc->cpuid,
+ sizeof(struct cpuid));
+ kvm->arch.model.ibc = proc->ibc;
+ kvm_s390_apply_fac_list_mask((long unsigned *)proc->fac_list);
+ memcpy(kvm->arch.model.fac_list, proc->fac_list,
+ S390_ARCH_FAC_LIST_SIZE_BYTE);
+ mutex_unlock(&kvm->lock);
+ kfree(proc);
+
+ return 0;
+}
+
+static int kvm_s390_set_cpu_model(struct kvm *kvm, struct kvm_device_attr *attr)
+{
+ int ret = -ENXIO;
+
+ switch (attr->attr) {
+ case KVM_S390_VM_CPU_PROCESSOR:
+ ret = kvm_s390_set_processor(kvm, attr);
+ break;
+ }
+ return ret;
+}
+
+static int kvm_s390_get_processor(struct kvm *kvm, struct kvm_device_attr *attr)
+{
+ struct kvm_s390_vm_cpu_processor *proc;
+ int rc = 0;
+
+ proc = kzalloc(sizeof(*proc), GFP_KERNEL);
+ if (!proc) {
+ rc = -ENOMEM;
+ goto out;
+ }
+ memcpy(&proc->cpuid, &kvm->arch.model.cpu_id, sizeof(struct cpuid));
+ proc->ibc = kvm->arch.model.ibc;
+ memcpy(&proc->fac_list, kvm->arch.model.fac_list,
+ S390_ARCH_FAC_LIST_SIZE_BYTE);
+ if (copy_to_user((void __user *)attr->addr, proc, sizeof(*proc)))
+ rc = -EFAULT;
+ kfree(proc);
+out:
+ return rc;
+}
+
+static int kvm_s390_get_machine(struct kvm *kvm, struct kvm_device_attr *attr)
+{
+ struct kvm_s390_vm_cpu_machine *mach;
+ int rc = 0;
+
+ mach = kzalloc(sizeof(*mach), GFP_KERNEL);
+ if (!mach) {
+ rc = -ENOMEM;
+ goto out;
+ }
+ get_cpu_id((struct cpuid *) &mach->cpuid);
+ mach->ibc_range = kvm_s390_lowest_ibc() << 16;
+ mach->ibc_range |= kvm_s390_latest_ibc();
+ memcpy(&mach->fac_mask, kvm_s390_fac_list_mask,
+ kvm_s390_fac_list_mask_size() * sizeof(u64));
+ kvm_s390_get_hard_fac_list((long unsigned int *) &mach->hard_fac_list,
+ S390_ARCH_FAC_LIST_SIZE_U64);
+ kvm_s390_get_soft_fac_list((long unsigned int *) &mach->soft_fac_list,
+ S390_ARCH_FAC_LIST_SIZE_U64);
I really have a hard time grasping what hard and soft means.
Hard facilities are those that are implemented by the CPU itself, either through processor
logic or be means of firmware micro code. That's the list returned by the STFL/STFLE
instruction. In addition to that, one can imagine that in future some of that features are
emulated on KVM side. These will be placed in the soft facility list and are optionally to
request by user space.
I don't see why we would have to differentiate between the two. User
space wants features enabled. Whether they are done in hardware or in
software doesn't matter.
I've tried to make my point on that in last answer of patch 3/6. It's a mistake
to think that user space just wants to have features, they come with different
qualities!
So? If I want to run a z9 compatible guest, I do -cpu z9. I can either

a) run it with emulation of a facility or
b) not run it

which one would the user choose?
If you run on a z990 host, you better use -cpu z990 because emulating some
fancy delta feature just cost additional CPU time. If the host is newer, please
go with -cpu z9.
Yes, I agree on that statement. Imagine a feature gets *dropped* though.
In that case -cpu z9 should enable emulation of that feature to maintain
migratability with a real z9 machine on newer hardware.
Nice try, but think what's happening in real world. Let's assume the feature is
TE again, available since zEC12 but would go away with zNext. In that case the
CPU model zNext-GA1 and all successors will not have zEC12 as supported model.
The application will just not run on that model if it insists on executing TE
instructions.

So what's the point in software emulated features then? Either we can emulate a feature or we can't. If we can, we can be compatible. If we can't, we're not compatible.


What user and thus also user space wants depends on other factors:

1. reliability
2. performance
3. availability

It's not features, that's what programmers want.

That's why I have designed the model and migration capability around the hardware
and not around the software features and don't allow them to be enabled currently
together.

A software feature is a nice add on that is helpful for evaluation or development
purpose. There is few space for it on productions systems.

One option that I currently see to make software implemented facility migration
capable is to calculate some kind of hash value derived from the full set of
active software facilities. That value can be compared with pre-calculated
values also stored in the supported model table of qemu. This value could be
seen like a virtual model extension that has to match like the model name.

But I have said it elsewhere already, a soft facility should be an exception and
not the rule.

So all we need is a list of "features the guest sees available" which is
the same as "features user space wants the guest to see" which then gets
masked through "features the host can do in hardware".

For emulation we can just check on the global feature availability on
whether we should emulate them or not.

Also, if user space wants to make sure that its feature list is actually
workable on the host kernel, it needs to set and get the features again
and then compare that with the ones it set? That's different from x86's
cpuid implementation but probably workable.
User space will probe what facilities are available and match them with the predefined cpu
model set. Only those models which use a partial or full subset of the hard/host facility
list are selectable.
Why?
If a host does not offer the features required for a model it is not able to
run efficiently.

Please take a look at how x86 does cpuid masking :).

In fact, I'm not 100% convinced that it's a good idea to link cpuid /
feature list exposure to the guest and actual feature implementation
inside the guest together. On POWER there is a patch set pending that
implements these two things separately - admittedly mostly because
hardware sucks and we can't change the PVR.
That is maybe the big difference with s390. The cpuid in the S390 case is not
directly comparable with the processor version register of POWER.

In the S390 world we have a well defined CPU model room spanned by the machine
type and its GA count. Thus we can define a bijective mapping between
(type, ga) <-> (cpuid, ibc, facility set). From type and ga we form the model
name which BTW is meaningful also for a human user.
Same thing as POWER.

By means of this name, a management interface (libvirt) will draw decisions if
migration to a remote hypervisor is a good idea or not. For that it just needs
to compare if the current model of the guest on the source hypervisor
("query-cpu-model"), is contained in the supported model list of the target
hypervisor ("query-cpu-definitions").
I don't think this works, since QEMU should always return all the cpu
definitions it's aware of on query-cpu-definitions, not just the ones
that it thinks may be compatible with the host at a random point in time.
It does not return model names that it thinks they are compatible at some point
in time. In s390 mode, it returns all definitions (CPU models) that a given host
system is capable to run. Together with the CPU model run by the guest, some upper
management interface knows if the hypervisor supports the required CPU model and
uses a guest definition with the same CPU model on the target hypervisor.

The information for that is taken from the model table which QEMU builds up during
startup time. This list limits the command line selectable CPU models as well.
This makes s390 derive from the way x86 handles things. NAK.
One second, that goes a little fast here :-). x86 returns a list they support which happens to
be the full list they define and s390 does logically the same because we know that certain
models are not supported due to probing. BTW that happens only if you run Qemu on back
level hardware and that is perfectly correct.

It's not what other architectures do and I'd hate to see s390 deviate just because.

The migration compatibility test is pretty much ARCH dependent. I looked into the
libvirt implementation and as one can see every architecture has its own implementation
there (libvirt/src/cpu/cpu_<arch>.c).

So here's my question again. How does x86 evaluate whether a target machine is compatible with a source machine?


Alex

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/