Skip to contentSkip to conversation

KVM: PPC: Book3S HV: add XIVE native exploitation mode

PATCH In review Cédric Le Goater 3 participants 26 messages Apr 18, 2019 – May 6, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v6 cover KVM: PPC: Book3S HV: add XIVE native exploitation mode

On the POWER9 processor, the XIVE interrupt controller can control interrupt sources using MMIOs to trigger events, to EOI or to turn off the sources. Priority management and interrupt acknowledgment is also controlled by MMIO in the CPU presenter sub-engine.

PowerNV/baremetal Linux runs natively under XIVE but sPAPR guests need special support from the hypervisor to do the same. This is called the XIVE native exploitation mode and today, it can be activated under the PowerPC Hypervisor, pHyp. However, Linux/KVM lacks XIVE native support and still offers the old interrupt mode interface using a KVM device implementing the XICS hcalls over XIVE.

The following series is proposal to add the same support under KVM.

A new KVM device is introduced for the XIVE native exploitation mode. It reuses most of the XICS-over-XIVE glue implementation structures which are internal to KVM but has a completely different interface. A set of KVM device ioctls provide support for the hypervisor calls, all handled in QEMU, to configure the sources and the event queues. From there, all interrupt control is transferred to the guest which can use MMIOs.

These MMIO regions (ESB and TIMA) are exposed to guests in QEMU, similarly to VFIO, and the associated VMAs are populated dynamically with the appropriate pages using a fault handler. These are now implemented using mmap()s of the KVM device fd.

Migration has its own specific needs regarding memory. The patchset provides a specific control to quiesce XIVE before capturing the memory. The save and restore of the internal state is based on the same ioctls used for the hcalls.

On a POWER9 sPAPR machine, the Client Architecture Support (CAS) negotiation process determines whether the guest operates with a interrupt controller using the XICS legacy model, as found on POWER8, or in XIVE exploitation mode. Which means that the KVM interrupt device should be created at run-time, after the machine has started. This requires extra support from KVM to destroy KVM devices. It is introduced at the end of the patchset and requires some attention.

This is based on Linux 5.1-rc5 and is a candidate for 5.2. The OPAL patches have been merged now.

GitHub trees available here :

QEMU sPAPR:

https://github.com/legoater/qemu/commits/xive-next

Linux/KVM:

https://github.com/legoater/linux/commits/xive-5.1

Thanks,

C.

Possible future todos :

- We should introduce a set of definitions common to the XIVE and XICS-on-XIVE devices
- The XICS-on-XIVE device file book3s_xive.c could be renamed to book3s_xics_on_xive.c or book3s_xics_p9.c
- The XICS-on-XIVE device still has some locking issues.
- xc->valid is not useful
- xc->xive is not very useful either. only used in xive_provision_queue()

- The XICS-on-XIVE device provisions XIVE resources "automagically" and QEMU needs to follow the same execution path regarding KVM else resources are not allocated correctly. This is a bit fragile and needs clarification. Compatibility might be complex to handle

Changes since v5:

- replaced the kvmppc_xive device array with named kvmppc_xive structs.
- in kvmppc_xive_cleanup_vcpu(), replaced xc->xive by vcpu->kvm->arch.xive which is always valid in this context
- in kvmppc_xive_release(), removed the comment on the race. The device being destroyed when the fd is closed, dev->private should always be valid when used. Also removed the kick_all_cpus_sync() to flush the IPIs.
- in the both release operation, removed the test on &kvm->online_vcpus.
- devices are now freed when the machine exits.

Changes since v4:

- add check on EQ page alignment
- add requirement on KVM_XIVE_EQ_ALWAYS_NOTIFY
- add documentation in Documentation/virtual/kvm/api.txt
- remove 'destroy' method
- introduce a 'release' device operation called when device fd is closed.
- introduce a 'xive_devices' array under the VM to store kvmppc_xive objects until VM is destroyed.

Changes since v3:

- removed a couple of useless includes
- fix the test ont the initial setting of the EQ toggle bit : 0 -> 1
- renamed qsize to qshift
- renamed qpage to qaddr
- checked host page size
- limited flags to KVM_XIVE_EQ_ALWAYS_NOTIFY to fit sPAPR specs
- Fixed xive_timaval description in documentation

Changes since v2:

- removed extra OPAL call definitions
- removed ->q_order setting. Only useful in the XICS-on-XIVE KVM device which allocates the EQs on behalf of the guest.
- returned -ENXIO when VP base is invalid
- made use of the xive_vp() macro to compute VP identifiers
- reworked locking in kvmppc_xive_native_connect_vcpu() to fix races
- stop advertising KVM_CAP_PPC_IRQ_XIVE as support is not fully available yet
- fixed comment on XIVE IRQ number space
- removed usage of the __x_* macros
- fixed locking on source block
- fixed comments on the KVM device attribute definitions
- handled MASKED EAS configuration
- fixed check on supported EQ size to restrict to 64K pages
- checked kvm_eq.flags that need to be zero
- removed the OPAL call when EQ qtoggle bit and index are zero.
- reduced the size of kvmppc_one_reg timaval attribute to two u64s
- stopped returning of the OS CAM line value

Changes since v1:

- Better documentation (was missing)
- Nested support. XIVE not advertised on non PowerNV platforms. This is a good way to test the fallback on QEMU emulated devices.
- ESB and TIMA special mapping done using the KVM device fd
- All hcalls moved to QEMU. Dropped the patch moving the hcall flags.
- Reworked of the KVM device ioctl controls to support hcalls and migration needs to capture/save states
- Merged the control syncing XIVE and marking the EQ page dirty
- Fixed passthrough support using the KVM device file address_space to clear the ESB pages from the mapping
- Misc enhancements and fixes

Cédric Le Goater (17):
powerpc/xive: add OPAL extensions for the XIVE native exploitation
support
KVM: PPC: Book3S HV: add a new KVM device for the XIVE native
exploitation mode
KVM: PPC: Book3S HV: XIVE: introduce a new capability
KVM_CAP_PPC_IRQ_XIVE
KVM: PPC: Book3S HV: XIVE: add a control to initialize a source
KVM: PPC: Book3S HV: XIVE: add a control to configure a source
KVM: PPC: Book3S HV: XIVE: add controls for the EQ configuration
KVM: PPC: Book3S HV: XIVE: add a global reset control
KVM: PPC: Book3S HV: XIVE: add a control to sync the sources
KVM: PPC: Book3S HV: XIVE: add a control to dirty the XIVE EQ pages
KVM: PPC: Book3S HV: XIVE: add get/set accessors for the VP XIVE state
KVM: introduce a 'mmap' method for KVM devices
KVM: PPC: Book3S HV: XIVE: add a TIMA mapping
KVM: PPC: Book3S HV: XIVE: add a mapping for the source ESB pages
KVM: PPC: Book3S HV: XIVE: add passthrough support
KVM: PPC: Book3S HV: XIVE: activate XIVE exploitation mode
KVM: introduce a 'release' method for KVM devices
KVM: PPC: Book3S HV: XIVE: replace the 'destroy' method by a 'release'
method

 arch/powerpc/include/asm/kvm_host.h        |    8 +-
 arch/powerpc/include/asm/kvm_ppc.h         |   32 +
 arch/powerpc/include/asm/opal-api.h        |    7 +-
 arch/powerpc/include/asm/opal.h            |    7 +
 arch/powerpc/include/asm/xive.h            |   17 +
 arch/powerpc/include/uapi/asm/kvm.h        |   46 +
 arch/powerpc/kvm/book3s_xive.h             |   37 +
 include/linux/kvm_host.h                   |   10 +
 include/uapi/linux/kvm.h                   |    3 +
 arch/powerpc/kvm/book3s.c                  |   31 +-
 arch/powerpc/kvm/book3s_xive.c             |  198 +++-
 arch/powerpc/kvm/book3s_xive_native.c      | 1218 ++++++++++++++++++++
 arch/powerpc/kvm/powerpc.c                 |   46 +
 arch/powerpc/platforms/powernv/opal-call.c |    3 +
 arch/powerpc/sysdev/xive/native.c          |  110 ++
 virt/kvm/kvm_main.c                        |   24 +
 Documentation/virtual/kvm/api.txt          |   10 +
 Documentation/virtual/kvm/devices/xive.txt |  197 ++++
 arch/powerpc/kvm/Makefile                  |    2 +-
 19 files changed, 1941 insertions(+), 65 deletions(-)
 create mode 100644 arch/powerpc/kvm/book3s_xive_native.c
 create mode 100644 Documentation/virtual/kvm/devices/xive.txt
Signature · 1 line
2.20.1
#1 lore raw

PATCH v6 1/17 powerpc/xive: add OPAL extensions for the XIVE native exploitation support

The support for XIVE native exploitation mode in Linux/KVM needs a couple more OPAL calls to get and set the state of the XIVE internal structures being used by a sPAPR guest.

---
 arch/powerpc/include/asm/opal-api.h        |  7 +-
 arch/powerpc/include/asm/opal.h            |  7 ++
 arch/powerpc/include/asm/xive.h            | 14 +++
 arch/powerpc/platforms/powernv/opal-call.c |  3 +
 arch/powerpc/sysdev/xive/native.c          | 99 ++++++++++++++++++++++
 5 files changed, 127 insertions(+), 3 deletions(-)
arch/powerpc/include/asm/opal-api.h +4 -3
index 870fb7b239ea..e1d118ac61dc 100644
@@ -186,8 +186,8 @@
 #define OPAL_XIVE_FREE_IRQ			140
 #define OPAL_XIVE_SYNC				141
 #define OPAL_XIVE_DUMP				142
-#define OPAL_XIVE_RESERVED3			143
-#define OPAL_XIVE_RESERVED4			144
+#define OPAL_XIVE_GET_QUEUE_STATE		143
+#define OPAL_XIVE_SET_QUEUE_STATE		144
 #define OPAL_SIGNAL_SYSTEM_RESET		145
 #define OPAL_NPU_INIT_CONTEXT			146
 #define OPAL_NPU_DESTROY_CONTEXT		147
@@ -210,7 +210,8 @@
 #define OPAL_PCI_GET_PBCQ_TUNNEL_BAR		164
 #define OPAL_PCI_SET_PBCQ_TUNNEL_BAR		165
 #define	OPAL_NX_COPROC_INIT			167
-#define OPAL_LAST				167
+#define OPAL_XIVE_GET_VP_STATE			170
+#define OPAL_LAST				170
 
 #define QUIESCE_HOLD			1 /* Spin all calls at entry */
 #define QUIESCE_REJECT			2 /* Fail all calls with OPAL_BUSY */
arch/powerpc/include/asm/opal.h +7 -0
index a55b01c90bb1..4e978d4dea5c 100644
@@ -279,6 +279,13 @@ int64_t opal_xive_allocate_irq(uint32_t chip_id);
 int64_t opal_xive_free_irq(uint32_t girq);
 int64_t opal_xive_sync(uint32_t type, uint32_t id);
 int64_t opal_xive_dump(uint32_t type, uint32_t id);
+int64_t opal_xive_get_queue_state(uint64_t vp, uint32_t prio,
+				  __be32 *out_qtoggle,
+				  __be32 *out_qindex);
+int64_t opal_xive_set_queue_state(uint64_t vp, uint32_t prio,
+				  uint32_t qtoggle,
+				  uint32_t qindex);
+int64_t opal_xive_get_vp_state(uint64_t vp, __be64 *out_w01);
 int64_t opal_pci_set_p2p(uint64_t phb_init, uint64_t phb_target,
 			uint64_t desc, uint16_t pe_number);
 
arch/powerpc/include/asm/xive.h +14 -0
index 3c704f5dd3ae..b579a943407b 100644
@@ -109,12 +109,26 @@ extern int xive_native_configure_queue(u32 vp_id, struct xive_q *q, u8 prio,
 extern void xive_native_disable_queue(u32 vp_id, struct xive_q *q, u8 prio);
 
 extern void xive_native_sync_source(u32 hw_irq);
+extern void xive_native_sync_queue(u32 hw_irq);
 extern bool is_xive_irq(struct irq_chip *chip);
 extern int xive_native_enable_vp(u32 vp_id, bool single_escalation);
 extern int xive_native_disable_vp(u32 vp_id);
 extern int xive_native_get_vp_info(u32 vp_id, u32 *out_cam_id, u32 *out_chip_id);
 extern bool xive_native_has_single_escalation(void);
 
+extern int xive_native_get_queue_info(u32 vp_id, uint32_t prio,
+				      u64 *out_qpage,
+				      u64 *out_qsize,
+				      u64 *out_qeoi_page,
+				      u32 *out_escalate_irq,
+				      u64 *out_qflags);
+
+extern int xive_native_get_queue_state(u32 vp_id, uint32_t prio, u32 *qtoggle,
+				       u32 *qindex);
+extern int xive_native_set_queue_state(u32 vp_id, uint32_t prio, u32 qtoggle,
+				       u32 qindex);
+extern int xive_native_get_vp_state(u32 vp_id, u64 *out_state);
+
 #else
 
 static inline bool xive_enabled(void) { return false; }
arch/powerpc/platforms/powernv/opal-call.c +3 -0
index daad8c45c8e7..7472244e7f30 100644
@@ -260,6 +260,9 @@ OPAL_CALL(opal_xive_get_vp_info,		OPAL_XIVE_GET_VP_INFO);
 OPAL_CALL(opal_xive_set_vp_info,		OPAL_XIVE_SET_VP_INFO);
 OPAL_CALL(opal_xive_sync,			OPAL_XIVE_SYNC);
 OPAL_CALL(opal_xive_dump,			OPAL_XIVE_DUMP);
+OPAL_CALL(opal_xive_get_queue_state,		OPAL_XIVE_GET_QUEUE_STATE);
+OPAL_CALL(opal_xive_set_queue_state,		OPAL_XIVE_SET_QUEUE_STATE);
+OPAL_CALL(opal_xive_get_vp_state,		OPAL_XIVE_GET_VP_STATE);
 OPAL_CALL(opal_signal_system_reset,		OPAL_SIGNAL_SYSTEM_RESET);
 OPAL_CALL(opal_npu_init_context,		OPAL_NPU_INIT_CONTEXT);
 OPAL_CALL(opal_npu_destroy_context,		OPAL_NPU_DESTROY_CONTEXT);
arch/powerpc/sysdev/xive/native.c +99 -0

108 lines not on this page · raw message

Signature · 1 line
2.20.1
#2 lore raw

PATCH v6 2/17 KVM: PPC: Book3S HV: add a new KVM device for the XIVE native exploitation mode

In reply to Cédric Le Goater

This is the basic framework for the new KVM device supporting the XIVE native exploitation mode. The user interface exposes a new KVM device to be created by QEMU, only available when running on a L0 hypervisor. Support for nested guests is not available yet.

The XIVE device reuses the device structure of the XICS-on-XIVE device as they have a lot in common. That could possibly change in the future if the need arise.

---
 arch/powerpc/include/asm/kvm_host.h        |   1 +
 arch/powerpc/include/asm/kvm_ppc.h         |   8 +
 arch/powerpc/include/uapi/asm/kvm.h        |   3 +
 include/uapi/linux/kvm.h                   |   2 +
 arch/powerpc/kvm/book3s.c                  |   7 +-
 arch/powerpc/kvm/book3s_xive_native.c      | 179 +++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt |  19 +++
 arch/powerpc/kvm/Makefile                  |   2 +-
 8 files changed, 219 insertions(+), 2 deletions(-)
 create mode 100644 arch/powerpc/kvm/book3s_xive_native.c
 create mode 100644 Documentation/virtual/kvm/devices/xive.txt
arch/powerpc/include/asm/kvm_host.h +1 -0
index e6b5bb012ccb..008523224e7a 100644
@@ -222,6 +222,7 @@ extern struct kvm_device_ops kvm_xics_ops;
 struct kvmppc_xive;
 struct kvmppc_xive_vcpu;
 extern struct kvm_device_ops kvm_xive_ops;
+extern struct kvm_device_ops kvm_xive_native_ops;
 
 struct kvmppc_passthru_irqmap;
 
arch/powerpc/include/asm/kvm_ppc.h +8 -0
index ac22b28ae78d..f3383e76017a 100644
@@ -594,6 +594,10 @@ extern int kvmppc_xive_set_icp(struct kvm_vcpu *vcpu, u64 icpval);
 extern int kvmppc_xive_set_irq(struct kvm *kvm, int irq_source_id, u32 irq,
 			       int level, bool line_status);
 extern void kvmppc_xive_push_vcpu(struct kvm_vcpu *vcpu);
+
+extern void kvmppc_xive_native_init_module(void);
+extern void kvmppc_xive_native_exit_module(void);
+
 #else
 static inline int kvmppc_xive_set_xive(struct kvm *kvm, u32 irq, u32 server,
 				       u32 priority) { return -1; }
@@ -617,6 +621,10 @@ static inline int kvmppc_xive_set_icp(struct kvm_vcpu *vcpu, u64 icpval) { retur
 static inline int kvmppc_xive_set_irq(struct kvm *kvm, int irq_source_id, u32 irq,
 				      int level, bool line_status) { return -ENODEV; }
 static inline void kvmppc_xive_push_vcpu(struct kvm_vcpu *vcpu) { }
+
+static inline void kvmppc_xive_native_init_module(void) { }
+static inline void kvmppc_xive_native_exit_module(void) { }
+
 #endif /* CONFIG_KVM_XIVE */
 
 #if defined(CONFIG_PPC_POWERNV) && defined(CONFIG_KVM_BOOK3S_64_HANDLER)
arch/powerpc/include/uapi/asm/kvm.h +3 -0
index 26ca425f4c2c..be0ce1f17625 100644
@@ -677,4 +677,7 @@ struct kvm_ppc_cpu_char {
 #define  KVM_XICS_PRESENTED		(1ULL << 43)
 #define  KVM_XICS_QUEUED		(1ULL << 44)
 
+/* POWER9 XIVE Native Interrupt Controller */
+#define KVM_DEV_XIVE_GRP_CTRL		1
+
 #endif /* __LINUX_KVM_POWERPC_H */
include/uapi/linux/kvm.h +2 -0
index 6d4ea4b6c922..e6368163d3a0 100644
@@ -1211,6 +1211,8 @@ enum kvm_device_type {
 #define KVM_DEV_TYPE_ARM_VGIC_V3	KVM_DEV_TYPE_ARM_VGIC_V3
 	KVM_DEV_TYPE_ARM_VGIC_ITS,
 #define KVM_DEV_TYPE_ARM_VGIC_ITS	KVM_DEV_TYPE_ARM_VGIC_ITS
+	KVM_DEV_TYPE_XIVE,
+#define KVM_DEV_TYPE_XIVE		KVM_DEV_TYPE_XIVE
 	KVM_DEV_TYPE_MAX,
 };
 
arch/powerpc/kvm/book3s.c +6 -1
index 10c5579d20ce..7c3348fa27e1 100644
@@ -1050,6 +1050,9 @@ static int kvmppc_book3s_init(void)
 	if (xics_on_xive()) {
 		kvmppc_xive_init_module();
 		kvm_register_device_ops(&kvm_xive_ops, KVM_DEV_TYPE_XICS);
+		kvmppc_xive_native_init_module();
+		kvm_register_device_ops(&kvm_xive_native_ops,
+					KVM_DEV_TYPE_XIVE);
 	} else
 #endif
 		kvm_register_device_ops(&kvm_xics_ops, KVM_DEV_TYPE_XICS);
@@ -1060,8 +1063,10 @@ static int kvmppc_book3s_init(void)
 static void kvmppc_book3s_exit(void)
 {
 #ifdef CONFIG_KVM_XICS
-	if (xics_on_xive())
+	if (xics_on_xive()) {
 		kvmppc_xive_exit_module();
+		kvmppc_xive_native_exit_module();
+	}
 #endif
 #ifdef CONFIG_KVM_BOOK3S_32_HANDLER
 	kvmppc_book3s_exit_pr();
arch/powerpc/kvm/book3s_xive_native.c new file +179 -0

179 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt new file +19 -0
new file mode 100644
index 000000000000..fdbd2ff92a88
@@ -0,0 +1,19 @@
+POWER9 eXternal Interrupt Virtualization Engine (XIVE Gen1)
+=============================
+
+Device types supported:
+  KVM_DEV_TYPE_XIVE     POWER9 XIVE Interrupt Controller generation 1
+
+This device acts as a VM interrupt controller. It provides the KVM
+interface to configure the interrupt sources of a VM in the underlying
+POWER9 XIVE interrupt controller.
+
+Only one XIVE instance may be instantiated. A guest XIVE device
+requires a POWER9 host and the guest OS should have support for the
+XIVE native exploitation interrupt mode. If not, it should run using
+the legacy interrupt mode, referred as XICS (POWER7/8).
+
+* Groups:
+
+  1. KVM_DEV_XIVE_GRP_CTRL
+  Provides global controls on the device
arch/powerpc/kvm/Makefile +1 -1
index 3223aec88b2c..4c67cc79de7c 100644
@@ -94,7 +94,7 @@ endif
 kvm-book3s_64-objs-$(CONFIG_KVM_XICS) += \
 	book3s_xics.o
 
-kvm-book3s_64-objs-$(CONFIG_KVM_XIVE) += book3s_xive.o
+kvm-book3s_64-objs-$(CONFIG_KVM_XIVE) += book3s_xive.o book3s_xive_native.o
 kvm-book3s_64-objs-$(CONFIG_SPAPR_TCE_IOMMU) += book3s_64_vio.o
 
 kvm-book3s_64-module-objs := \
Signature · 1 line
2.20.1
#3 lore raw

PATCH v6 3/17 KVM: PPC: Book3S HV: XIVE: introduce a new capability KVM_CAP_PPC_IRQ_XIVE

In reply to Cédric Le Goater

The user interface exposes a new capability KVM_CAP_PPC_IRQ_XIVE to let QEMU connect the vCPU presenters to the XIVE KVM device if required. The capability is not advertised for now as the full support for the XIVE native exploitation mode is not yet available. When this is case, the capability will be advertised on PowerNV Hypervisors only. Nested guests (pseries KVM Hypervisor) are not supported.

Internally, the interface to the new KVM device is protected with a new interrupt mode: KVMPPC_IRQ_XIVE.

---
 arch/powerpc/include/asm/kvm_host.h   |   1 +
 arch/powerpc/include/asm/kvm_ppc.h    |  13 +++
 arch/powerpc/kvm/book3s_xive.h        |  11 ++
 include/uapi/linux/kvm.h              |   1 +
 arch/powerpc/kvm/book3s_xive.c        |  88 ++++++++-------
 arch/powerpc/kvm/book3s_xive_native.c | 150 ++++++++++++++++++++++++++
 arch/powerpc/kvm/powerpc.c            |  36 +++++++
 Documentation/virtual/kvm/api.txt     |   9 ++
 8 files changed, 268 insertions(+), 41 deletions(-)
arch/powerpc/include/asm/kvm_host.h +1 -0
index 008523224e7a..9cc6abdce1b9 100644
@@ -450,6 +450,7 @@ struct kvmppc_passthru_irqmap {
 #define KVMPPC_IRQ_DEFAULT	0
 #define KVMPPC_IRQ_MPIC		1
 #define KVMPPC_IRQ_XICS		2 /* Includes a XIVE option */
+#define KVMPPC_IRQ_XIVE		3 /* XIVE native exploitation mode */
 
 #define MMIO_HPTE_CACHE_SIZE	4
 
arch/powerpc/include/asm/kvm_ppc.h +13 -0
index f3383e76017a..6928a35ac3c7 100644
@@ -595,6 +595,14 @@ extern int kvmppc_xive_set_irq(struct kvm *kvm, int irq_source_id, u32 irq,
 			       int level, bool line_status);
 extern void kvmppc_xive_push_vcpu(struct kvm_vcpu *vcpu);
 
+static inline int kvmppc_xive_enabled(struct kvm_vcpu *vcpu)
+{
+	return vcpu->arch.irq_type = KVMPPC_IRQ_XIVE;
+}
+
+extern int kvmppc_xive_native_connect_vcpu(struct kvm_device *dev,
+					   struct kvm_vcpu *vcpu, u32 cpu);
+extern void kvmppc_xive_native_cleanup_vcpu(struct kvm_vcpu *vcpu);
 extern void kvmppc_xive_native_init_module(void);
 extern void kvmppc_xive_native_exit_module(void);
 
@@ -622,6 +630,11 @@ static inline int kvmppc_xive_set_irq(struct kvm *kvm, int irq_source_id, u32 ir
 				      int level, bool line_status) { return -ENODEV; }
 static inline void kvmppc_xive_push_vcpu(struct kvm_vcpu *vcpu) { }
 
+static inline int kvmppc_xive_enabled(struct kvm_vcpu *vcpu)
+	{ return 0; }
+static inline int kvmppc_xive_native_connect_vcpu(struct kvm_device *dev,
+			  struct kvm_vcpu *vcpu, u32 cpu) { return -EBUSY; }
+static inline void kvmppc_xive_native_cleanup_vcpu(struct kvm_vcpu *vcpu) { }
 static inline void kvmppc_xive_native_init_module(void) { }
 static inline void kvmppc_xive_native_exit_module(void) { }
 
arch/powerpc/kvm/book3s_xive.h +11 -0
index a08ae6fd4c51..d366df69b9cb 100644
@@ -198,6 +198,11 @@ static inline struct kvmppc_xive_src_block *kvmppc_xive_find_source(struct kvmpp
 	return xive->src_blocks[bid];
 }
 
+static inline u32 kvmppc_xive_vp(struct kvmppc_xive *xive, u32 server)
+{
+	return xive->vp_base + kvmppc_pack_vcpu_id(xive->kvm, server);
+}
+
 /*
  * Mapping between guest priorities and host priorities
  * is as follow.
@@ -248,5 +253,11 @@ extern int (*__xive_vm_h_ipi)(struct kvm_vcpu *vcpu, unsigned long server,
 extern int (*__xive_vm_h_cppr)(struct kvm_vcpu *vcpu, unsigned long cppr);
 extern int (*__xive_vm_h_eoi)(struct kvm_vcpu *vcpu, unsigned long xirr);
 
+/*
+ * Common Xive routines for XICS-over-XIVE and XIVE native
+ */
+void kvmppc_xive_disable_vcpu_interrupts(struct kvm_vcpu *vcpu);
+int kvmppc_xive_debug_show_queues(struct seq_file *m, struct kvm_vcpu *vcpu);
+
 #endif /* CONFIG_KVM_XICS */
 #endif /* _KVM_PPC_BOOK3S_XICS_H */
include/uapi/linux/kvm.h +1 -0
index e6368163d3a0..52bf74a1616e 100644
@@ -988,6 +988,7 @@ struct kvm_ppc_resize_hpt {
 #define KVM_CAP_ARM_VM_IPA_SIZE 165
 #define KVM_CAP_MANUAL_DIRTY_LOG_PROTECT 166
 #define KVM_CAP_HYPERV_CPUID 167
+#define KVM_CAP_PPC_IRQ_XIVE 168
 
 #ifdef KVM_CAP_IRQ_ROUTING
 
arch/powerpc/kvm/book3s_xive.c +46 -41

154 lines not on this page · raw message

arch/powerpc/kvm/book3s_xive_native.c +150 -0

166 lines not on this page · raw message

arch/powerpc/kvm/powerpc.c +36 -0
index 8885377ec3e0..b0858ee61460 100644
@@ -570,6 +570,15 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 	case KVM_CAP_PPC_GET_CPU_CHAR:
 		r = 1;
 		break;
+#ifdef CONFIG_KVM_XIVE
+	case KVM_CAP_PPC_IRQ_XIVE:
+		/*
+		 * Return false until all the XIVE infrastructure is
+		 * in place including support for migration.
+		 */
+		r = 0;
+		break;
+#endif
 
 	case KVM_CAP_PPC_ALLOC_HTAB:
 		r = hv_enabled;
@@ -753,6 +762,9 @@ void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 		else
 			kvmppc_xics_free_icp(vcpu);
 		break;
+	case KVMPPC_IRQ_XIVE:
+		kvmppc_xive_native_cleanup_vcpu(vcpu);
+		break;
 	}
 
 	kvmppc_core_vcpu_free(vcpu);
@@ -1941,6 +1953,30 @@ static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
 		break;
 	}
 #endif /* CONFIG_KVM_XICS */
+#ifdef CONFIG_KVM_XIVE
+	case KVM_CAP_PPC_IRQ_XIVE: {
+		struct fd f;
+		struct kvm_device *dev;
+
+		r = -EBADF;
+		f = fdget(cap->args[0]);
+		if (!f.file)
+			break;
+
+		r = -ENXIO;
+		if (!xive_enabled())
+			break;
+
+		r = -EPERM;
+		dev = kvm_device_from_filp(f.file);
+		if (dev)
+			r = kvmppc_xive_native_connect_vcpu(dev, vcpu,
+							    cap->args[1]);
+
+		fdput(f);
+		break;
+	}
+#endif /* CONFIG_KVM_XIVE */
 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 	case KVM_CAP_PPC_FWNMI:
 		r = -EINVAL;
Documentation/virtual/kvm/api.txt +9 -0
index 67068c47c591..e38eb17b7be6 100644
@@ -4504,6 +4504,15 @@ struct kvm_sync_regs {
         struct kvm_vcpu_events events;
 };
 
+6.75 KVM_CAP_PPC_IRQ_XIVE
+
+Architectures: ppc
+Target: vcpu
+Parameters: args[0] is the XIVE device fd
+            args[1] is the XIVE CPU number (server ID) for this vcpu
+
+This capability connects the vcpu to an in-kernel XIVE device.
+
 7. Capabilities that can be enabled on VMs
 ------------------------------------------
 
Signature · 1 line
2.20.1
#4 lore raw

PATCH v6 4/17 KVM: PPC: Book3S HV: XIVE: add a control to initialize a source

In reply to Cédric Le Goater

The XIVE KVM device maintains a list of interrupt sources for the VM which are allocated in the pool of generic interrupts (IPIs) of the main XIVE IC controller. These are used for the CPU IPIs as well as for virtual device interrupts. The IRQ number space is defined by QEMU.

The XIVE device reuses the source structures of the XICS-on-XIVE device for the source blocks (2-level tree) and for the source interrupts. Under XIVE native, the source interrupt caches mostly configuration information and is less used than under the XICS-on-XIVE device in which hcalls are still necessary at run-time.

When a source is initialized in KVM, an IPI interrupt source is simply allocated at the OPAL level and then MASKED. KVM only needs to know about its type: LSI or MSI.

---
 arch/powerpc/include/uapi/asm/kvm.h        |   5 +
 arch/powerpc/kvm/book3s_xive.h             |  10 ++
 arch/powerpc/kvm/book3s_xive.c             |   8 +-
 arch/powerpc/kvm/book3s_xive_native.c      | 106 +++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt |  15 +++
 5 files changed, 140 insertions(+), 4 deletions(-)
arch/powerpc/include/uapi/asm/kvm.h +5 -0
index be0ce1f17625..d468294c2a67 100644
@@ -679,5 +679,10 @@ struct kvm_ppc_cpu_char {
 
 /* POWER9 XIVE Native Interrupt Controller */
 #define KVM_DEV_XIVE_GRP_CTRL		1
+#define KVM_DEV_XIVE_GRP_SOURCE		2	/* 64-bit source identifier */
+
+/* Layout of 64-bit XIVE source attribute values */
+#define KVM_XIVE_LEVEL_SENSITIVE	(1ULL << 0)
+#define KVM_XIVE_LEVEL_ASSERTED		(1ULL << 1)
 
 #endif /* __LINUX_KVM_POWERPC_H */
arch/powerpc/kvm/book3s_xive.h +10 -0
index d366df69b9cb..1be921cb5dcb 100644
@@ -12,6 +12,13 @@
 #ifdef CONFIG_KVM_XICS
 #include "book3s_xics.h"
 
+/*
+ * The XIVE Interrupt source numbers are within the range 0 to
+ * KVMPPC_XICS_NR_IRQS.
+ */
+#define KVMPPC_XIVE_FIRST_IRQ	0
+#define KVMPPC_XIVE_NR_IRQS	KVMPPC_XICS_NR_IRQS
+
 /*
  * State for one guest irq source.
  *
@@ -258,6 +265,9 @@ extern int (*__xive_vm_h_eoi)(struct kvm_vcpu *vcpu, unsigned long xirr);
  */
 void kvmppc_xive_disable_vcpu_interrupts(struct kvm_vcpu *vcpu);
 int kvmppc_xive_debug_show_queues(struct seq_file *m, struct kvm_vcpu *vcpu);
+struct kvmppc_xive_src_block *kvmppc_xive_create_src_block(
+	struct kvmppc_xive *xive, int irq);
+void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb);
 
 #endif /* CONFIG_KVM_XICS */
 #endif /* _KVM_PPC_BOOK3S_XICS_H */
arch/powerpc/kvm/book3s_xive.c +4 -4
index e7f1ada1c3de..6c9f9fd0855f 100644
@@ -1480,8 +1480,8 @@ static int xive_get_source(struct kvmppc_xive *xive, long irq, u64 addr)
 	return 0;
 }
 
-static struct kvmppc_xive_src_block *xive_create_src_block(struct kvmppc_xive *xive,
-							   int irq)
+struct kvmppc_xive_src_block *kvmppc_xive_create_src_block(
+	struct kvmppc_xive *xive, int irq)
 {
 	struct kvm *kvm = xive->kvm;
 	struct kvmppc_xive_src_block *sb;
@@ -1560,7 +1560,7 @@ static int xive_set_source(struct kvmppc_xive *xive, long irq, u64 addr)
 	sb = kvmppc_xive_find_source(xive, irq, &idx);
 	if (!sb) {
 		pr_devel("No source, creating source block...\n");
-		sb = xive_create_src_block(xive, irq);
+		sb = kvmppc_xive_create_src_block(xive, irq);
 		if (!sb) {
 			pr_devel("Failed to create block...\n");
 			return -ENOMEM;
@@ -1784,7 +1784,7 @@ static void kvmppc_xive_cleanup_irq(u32 hw_num, struct xive_irq_data *xd)
 	xive_cleanup_irq_data(xd);
 }
 
-static void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb)
+void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb)
 {
 	int i;
 
arch/powerpc/kvm/book3s_xive_native.c +106 -0

142 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +15 -0
index fdbd2ff92a88..cd8bfc37b72e 100644
@@ -17,3 +17,18 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
 
   1. KVM_DEV_XIVE_GRP_CTRL
   Provides global controls on the device
+
+  2. KVM_DEV_XIVE_GRP_SOURCE (write only)
+  Initializes a new source in the XIVE device and mask it.
+  Attributes:
+    Interrupt source number  (64-bit)
+  The kvm_device_attr.addr points to a __u64 value:
+  bits:     | 63   ....  2 |   1   |   0
+  values:   |    unused    | level | type
+  - type:  0:MSI 1:LSI
+  - level: assertion level in case of an LSI.
+  Errors:
+    -E2BIG:  Interrupt source number is out of range
+    -ENOMEM: Could not create a new source block
+    -EFAULT: Invalid user pointer for attr->addr.
+    -ENXIO:  Could not allocate underlying HW interrupt
Signature · 1 line
2.20.1
#5 lore raw

PATCH v6 5/17 KVM: PPC: Book3S HV: XIVE: add a control to configure a source

In reply to Cédric Le Goater

This control will be used by the H_INT_SET_SOURCE_CONFIG hcall from QEMU to configure the target of a source and also to restore the configuration of a source when migrating the VM.

The XIVE source interrupt structure is extended with the value of the Effective Interrupt Source Number. The EISN is the interrupt number pushed in the event queue that the guest OS will use to dispatch events internally. Caching the EISN value in KVM eases the test when checking if a reconfiguration is indeed needed.

---
 arch/powerpc/include/uapi/asm/kvm.h        | 11 +++
 arch/powerpc/kvm/book3s_xive.h             |  4 +
 arch/powerpc/kvm/book3s_xive.c             |  5 +-
 arch/powerpc/kvm/book3s_xive_native.c      | 97 ++++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt | 21 +++++
 5 files changed, 136 insertions(+), 2 deletions(-)
arch/powerpc/include/uapi/asm/kvm.h +11 -0
index d468294c2a67..e8161e21629b 100644
@@ -680,9 +680,20 @@ struct kvm_ppc_cpu_char {
 /* POWER9 XIVE Native Interrupt Controller */
 #define KVM_DEV_XIVE_GRP_CTRL		1
 #define KVM_DEV_XIVE_GRP_SOURCE		2	/* 64-bit source identifier */
+#define KVM_DEV_XIVE_GRP_SOURCE_CONFIG	3	/* 64-bit source identifier */
 
 /* Layout of 64-bit XIVE source attribute values */
 #define KVM_XIVE_LEVEL_SENSITIVE	(1ULL << 0)
 #define KVM_XIVE_LEVEL_ASSERTED		(1ULL << 1)
 
+/* Layout of 64-bit XIVE source configuration attribute values */
+#define KVM_XIVE_SOURCE_PRIORITY_SHIFT	0
+#define KVM_XIVE_SOURCE_PRIORITY_MASK	0x7
+#define KVM_XIVE_SOURCE_SERVER_SHIFT	3
+#define KVM_XIVE_SOURCE_SERVER_MASK	0xfffffff8ULL
+#define KVM_XIVE_SOURCE_MASKED_SHIFT	32
+#define KVM_XIVE_SOURCE_MASKED_MASK	0x100000000ULL
+#define KVM_XIVE_SOURCE_EISN_SHIFT	33
+#define KVM_XIVE_SOURCE_EISN_MASK	0xfffffffe00000000ULL
+
 #endif /* __LINUX_KVM_POWERPC_H */
arch/powerpc/kvm/book3s_xive.h +4 -0
index 1be921cb5dcb..ae26fe653d98 100644
@@ -61,6 +61,9 @@ struct kvmppc_xive_irq_state {
 	bool saved_p;
 	bool saved_q;
 	u8 saved_scan_prio;
+
+	/* Xive native */
+	u32 eisn;			/* Guest Effective IRQ number */
 };
 
 /* Select the "right" interrupt (IPI vs. passthrough) */
@@ -268,6 +271,7 @@ int kvmppc_xive_debug_show_queues(struct seq_file *m, struct kvm_vcpu *vcpu);
 struct kvmppc_xive_src_block *kvmppc_xive_create_src_block(
 	struct kvmppc_xive *xive, int irq);
 void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb);
+int kvmppc_xive_select_target(struct kvm *kvm, u32 *server, u8 prio);
 
 #endif /* CONFIG_KVM_XICS */
 #endif /* _KVM_PPC_BOOK3S_XICS_H */
arch/powerpc/kvm/book3s_xive.c +3 -2
index 6c9f9fd0855f..e09f3addffe5 100644
@@ -342,7 +342,7 @@ static int xive_try_pick_queue(struct kvm_vcpu *vcpu, u8 prio)
 	return atomic_add_unless(&q->count, 1, max) ? 0 : -EBUSY;
 }
 
-static int xive_select_target(struct kvm *kvm, u32 *server, u8 prio)
+int kvmppc_xive_select_target(struct kvm *kvm, u32 *server, u8 prio)
 {
 	struct kvm_vcpu *vcpu;
 	int i, rc;
@@ -530,7 +530,7 @@ static int xive_target_interrupt(struct kvm *kvm,
 	 * priority. The count for that new target will have
 	 * already been incremented.
 	 */
-	rc = xive_select_target(kvm, &server, prio);
+	rc = kvmppc_xive_select_target(kvm, &server, prio);
 
 	/*
 	 * We failed to find a target ? Not much we can do
@@ -1504,6 +1504,7 @@ struct kvmppc_xive_src_block *kvmppc_xive_create_src_block(
 
 	for (i = 0; i < KVMPPC_XICS_IRQ_PER_ICS; i++) {
 		sb->irq_state[i].number = (bid << KVMPPC_XICS_ICS_SHIFT) | i;
+		sb->irq_state[i].eisn = 0;
 		sb->irq_state[i].guest_priority = MASKED;
 		sb->irq_state[i].saved_priority = MASKED;
 		sb->irq_state[i].act_priority = MASKED;
arch/powerpc/kvm/book3s_xive_native.c +97 -0

115 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +21 -0
index cd8bfc37b72e..33c64b2cdbe8 100644
@@ -32,3 +32,24 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
     -ENOMEM: Could not create a new source block
     -EFAULT: Invalid user pointer for attr->addr.
     -ENXIO:  Could not allocate underlying HW interrupt
+
+  3. KVM_DEV_XIVE_GRP_SOURCE_CONFIG (write only)
+  Configures source targeting
+  Attributes:
+    Interrupt source number  (64-bit)
+  The kvm_device_attr.addr points to a __u64 value:
+  bits:     | 63   ....  33 |  32  | 31 .. 3 |  2 .. 0
+  values:   |    eisn       | mask |  server | priority
+  - priority: 0-7 interrupt priority level
+  - server: CPU number chosen to handle the interrupt
+  - mask: mask flag (unused)
+  - eisn: Effective Interrupt Source Number
+  Errors:
+    -ENOENT: Unknown source number
+    -EINVAL: Not initialized source number
+    -EINVAL: Invalid priority
+    -EINVAL: Invalid CPU number.
+    -EFAULT: Invalid user pointer for attr->addr.
+    -ENXIO:  CPU event queues not configured or configuration of the
+             underlying HW interrupt failed
+    -EBUSY:  No CPU available to serve interrupt
Signature · 1 line
2.20.1
#6 lore raw

PATCH v6 6/17 KVM: PPC: Book3S HV: XIVE: add controls for the EQ configuration

In reply to Cédric Le Goater

These controls will be used by the H_INT_SET_QUEUE_CONFIG and H_INT_GET_QUEUE_CONFIG hcalls from QEMU to configure the underlying Event Queue in the XIVE IC. They will also be used to restore the configuration of the XIVE EQs and to capture the internal run-time state of the EQs. Both 'get' and 'set' rely on an OPAL call to access the EQ toggle bit and EQ index which are updated by the XIVE IC when event notifications are enqueued in the EQ.

The value of the guest physical address of the event queue is saved in the XIVE internal xive_q structure for later use. That is when migration needs to mark the EQ pages dirty to capture a consistent memory state of the VM.

To be noted that H_INT_SET_QUEUE_CONFIG does not require the extra OPAL call setting the EQ toggle bit and EQ index to configure the EQ, but restoring the EQ state will.

---
 arch/powerpc/include/asm/xive.h            |   2 +
 arch/powerpc/include/uapi/asm/kvm.h        |  19 ++
 arch/powerpc/kvm/book3s_xive.h             |   2 +
 arch/powerpc/kvm/book3s_xive.c             |  15 +-
 arch/powerpc/kvm/book3s_xive_native.c      | 249 +++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt |  34 +++
 6 files changed, 315 insertions(+), 6 deletions(-)
arch/powerpc/include/asm/xive.h +2 -0
index b579a943407b..c4e88abd3b67 100644
@@ -73,6 +73,8 @@ struct xive_q {
 	u32			esc_irq;
 	atomic_t		count;
 	atomic_t		pending_count;
+	u64			guest_qaddr;
+	u32			guest_qshift;
 };
 
 /* Global enable flags for the XIVE support */
arch/powerpc/include/uapi/asm/kvm.h +19 -0
index e8161e21629b..85005400fd86 100644
@@ -681,6 +681,7 @@ struct kvm_ppc_cpu_char {
 #define KVM_DEV_XIVE_GRP_CTRL		1
 #define KVM_DEV_XIVE_GRP_SOURCE		2	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_SOURCE_CONFIG	3	/* 64-bit source identifier */
+#define KVM_DEV_XIVE_GRP_EQ_CONFIG	4	/* 64-bit EQ identifier */
 
 /* Layout of 64-bit XIVE source attribute values */
 #define KVM_XIVE_LEVEL_SENSITIVE	(1ULL << 0)
@@ -696,4 +697,22 @@ struct kvm_ppc_cpu_char {
 #define KVM_XIVE_SOURCE_EISN_SHIFT	33
 #define KVM_XIVE_SOURCE_EISN_MASK	0xfffffffe00000000ULL
 
+/* Layout of 64-bit EQ identifier */
+#define KVM_XIVE_EQ_PRIORITY_SHIFT	0
+#define KVM_XIVE_EQ_PRIORITY_MASK	0x7
+#define KVM_XIVE_EQ_SERVER_SHIFT	3
+#define KVM_XIVE_EQ_SERVER_MASK		0xfffffff8ULL
+
+/* Layout of EQ configuration values (64 bytes) */
+struct kvm_ppc_xive_eq {
+	__u32 flags;
+	__u32 qshift;
+	__u64 qaddr;
+	__u32 qtoggle;
+	__u32 qindex;
+	__u8  pad[40];
+};
+
+#define KVM_XIVE_EQ_ALWAYS_NOTIFY	0x00000001
+
 #endif /* __LINUX_KVM_POWERPC_H */
arch/powerpc/kvm/book3s_xive.h +2 -0
index ae26fe653d98..622f594d93e1 100644
@@ -272,6 +272,8 @@ struct kvmppc_xive_src_block *kvmppc_xive_create_src_block(
 	struct kvmppc_xive *xive, int irq);
 void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb);
 int kvmppc_xive_select_target(struct kvm *kvm, u32 *server, u8 prio);
+int kvmppc_xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio,
+				  bool single_escalation);
 
 #endif /* CONFIG_KVM_XICS */
 #endif /* _KVM_PPC_BOOK3S_XICS_H */
arch/powerpc/kvm/book3s_xive.c +9 -6
index e09f3addffe5..c1b7aa7dbc28 100644
@@ -166,7 +166,8 @@ static irqreturn_t xive_esc_irq(int irq, void *data)
 	return IRQ_HANDLED;
 }
 
-static int xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio)
+int kvmppc_xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio,
+				  bool single_escalation)
 {
 	struct kvmppc_xive_vcpu *xc = vcpu->arch.xive_vcpu;
 	struct xive_q *q = &xc->queues[prio];
@@ -185,7 +186,7 @@ static int xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio)
 		return -EIO;
 	}
 
-	if (xc->xive->single_escalation)
+	if (single_escalation)
 		name = kasprintf(GFP_KERNEL, "kvm-%d-%d",
 				 vcpu->kvm->arch.lpid, xc->server_num);
 	else
@@ -217,7 +218,7 @@ static int xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio)
 	 * interrupt, thus leaving it effectively masked after
 	 * it fires once.
 	 */
-	if (xc->xive->single_escalation) {
+	if (single_escalation) {
 		struct irq_data *d = irq_get_irq_data(xc->esc_virq[prio]);
 		struct xive_irq_data *xd = irq_data_get_irq_handler_data(d);
 
@@ -291,7 +292,8 @@ static int xive_check_provisioning(struct kvm *kvm, u8 prio)
 			continue;
 		rc = xive_provision_queue(vcpu, prio);
 		if (rc = 0 && !xive->single_escalation)
-			xive_attach_escalation(vcpu, prio);
+			kvmppc_xive_attach_escalation(vcpu, prio,
+						      xive->single_escalation);
 		if (rc)
 			return rc;
 	}
@@ -1214,7 +1216,8 @@ int kvmppc_xive_connect_vcpu(struct kvm_device *dev,
 		if (xive->qmap & (1 << i)) {
 			r = xive_provision_queue(vcpu, i);
 			if (r = 0 && !xive->single_escalation)
-				xive_attach_escalation(vcpu, i);
+				kvmppc_xive_attach_escalation(
+					vcpu, i, xive->single_escalation);
 			if (r)
 				goto bail;
 		} else {
@@ -1229,7 +1232,7 @@ int kvmppc_xive_connect_vcpu(struct kvm_device *dev,
 	}
 
 	/* If not done above, attach priority 0 escalation */
-	r = xive_attach_escalation(vcpu, 0);
+	r = kvmppc_xive_attach_escalation(vcpu, 0, xive->single_escalation);
 	if (r)
 		goto bail;
 
arch/powerpc/kvm/book3s_xive_native.c +249 -0

273 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +34 -0
index 33c64b2cdbe8..cc13bfd5cf53 100644
@@ -53,3 +53,37 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
     -ENXIO:  CPU event queues not configured or configuration of the
              underlying HW interrupt failed
     -EBUSY:  No CPU available to serve interrupt
+
+  4. KVM_DEV_XIVE_GRP_EQ_CONFIG (read-write)
+  Configures an event queue of a CPU
+  Attributes:
+    EQ descriptor identifier (64-bit)
+  The EQ descriptor identifier is a tuple (server, priority) :
+  bits:     | 63   ....  32 | 31 .. 3 |  2 .. 0
+  values:   |    unused     |  server | priority
+  The kvm_device_attr.addr points to :
+    struct kvm_ppc_xive_eq {
+	__u32 flags;
+	__u32 qshift;
+	__u64 qaddr;
+	__u32 qtoggle;
+	__u32 qindex;
+	__u8  pad[40];
+    };
+  - flags: queue flags
+    KVM_XIVE_EQ_ALWAYS_NOTIFY (required)
+	forces notification without using the coalescing mechanism
+	provided by the XIVE END ESBs.
+  - qshift: queue size (power of 2)
+  - qaddr: real address of queue
+  - qtoggle: current queue toggle bit
+  - qindex: current queue index
+  - pad: reserved for future use
+  Errors:
+    -ENOENT: Invalid CPU number
+    -EINVAL: Invalid priority
+    -EINVAL: Invalid flags
+    -EINVAL: Invalid queue size
+    -EINVAL: Invalid queue address
+    -EFAULT: Invalid user pointer for attr->addr.
+    -EIO:    Configuration of the underlying HW failed
Signature · 1 line
2.20.1
#7 lore raw

PATCH v6 7/17 KVM: PPC: Book3S HV: XIVE: add a global reset control

In reply to Cédric Le Goater

This control is to be used by the H_INT_RESET hcall from QEMU. Its purpose is to clear all configuration of the sources and EQs. This is necessary in case of a kexec (for a kdump kernel for instance) to make sure that no remaining configuration is left from the previous boot setup so that the new kernel can start safely from a clean state.

The queue 7 is ignored when the XIVE device is configured to run in single escalation mode. Prio 7 is used by escalations.

The XIVE VP is kept enabled as the vCPU is still active and connected to the XIVE device.

---
 arch/powerpc/include/uapi/asm/kvm.h        |  1 +
 arch/powerpc/kvm/book3s_xive_native.c      | 85 ++++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt |  5 ++
 3 files changed, 91 insertions(+)
arch/powerpc/include/uapi/asm/kvm.h +1 -0
index 85005400fd86..f045f9dee42e 100644
@@ -679,6 +679,7 @@ struct kvm_ppc_cpu_char {
 
 /* POWER9 XIVE Native Interrupt Controller */
 #define KVM_DEV_XIVE_GRP_CTRL		1
+#define   KVM_DEV_XIVE_RESET		1
 #define KVM_DEV_XIVE_GRP_SOURCE		2	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_SOURCE_CONFIG	3	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_EQ_CONFIG	4	/* 64-bit EQ identifier */
arch/powerpc/kvm/book3s_xive_native.c +85 -0

103 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +5 -0
index cc13bfd5cf53..429cbc4cf960 100644
@@ -17,6 +17,11 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
 
   1. KVM_DEV_XIVE_GRP_CTRL
   Provides global controls on the device
+  Attributes:
+    1.1 KVM_DEV_XIVE_RESET (write only)
+    Resets the interrupt controller configuration for sources and event
+    queues. To be used by kexec and kdump.
+    Errors: none
 
   2. KVM_DEV_XIVE_GRP_SOURCE (write only)
   Initializes a new source in the XIVE device and mask it.
Signature · 1 line
2.20.1
#8 lore raw

PATCH v6 8/17 KVM: PPC: Book3S HV: XIVE: add a control to sync the sources

In reply to Cédric Le Goater

This control will be used by the H_INT_SYNC hcall from QEMU to flush event notifications on the XIVE IC owning the source.

---
 arch/powerpc/include/uapi/asm/kvm.h        |  1 +
 arch/powerpc/kvm/book3s_xive_native.c      | 36 ++++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt |  8 +++++
 3 files changed, 45 insertions(+)
arch/powerpc/include/uapi/asm/kvm.h +1 -0
index f045f9dee42e..e4abe30f6fc6 100644
@@ -683,6 +683,7 @@ struct kvm_ppc_cpu_char {
 #define KVM_DEV_XIVE_GRP_SOURCE		2	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_SOURCE_CONFIG	3	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_EQ_CONFIG	4	/* 64-bit EQ identifier */
+#define KVM_DEV_XIVE_GRP_SOURCE_SYNC	5       /* 64-bit source identifier */
 
 /* Layout of 64-bit XIVE source attribute values */
 #define KVM_XIVE_LEVEL_SENSITIVE	(1ULL << 0)
arch/powerpc/kvm/book3s_xive_native.c +36 -0
index b9597d80c95a..65380416d101 100644
@@ -335,6 +335,38 @@ static int kvmppc_xive_native_set_source_config(struct kvmppc_xive *xive,
 						       priority, masked, eisn);
 }
 
+static int kvmppc_xive_native_sync_source(struct kvmppc_xive *xive,
+					  long irq, u64 addr)
+{
+	struct kvmppc_xive_src_block *sb;
+	struct kvmppc_xive_irq_state *state;
+	struct xive_irq_data *xd;
+	u32 hw_num;
+	u16 src;
+	int rc = 0;
+
+	pr_devel("%s irq=0x%lx", __func__, irq);
+
+	sb = kvmppc_xive_find_source(xive, irq, &src);
+	if (!sb)
+		return -ENOENT;
+
+	state = &sb->irq_state[src];
+
+	rc = -EINVAL;
+
+	arch_spin_lock(&sb->lock);
+
+	if (state->valid) {
+		kvmppc_xive_select_irq(state, &hw_num, &xd);
+		xive_native_sync_source(hw_num);
+		rc = 0;
+	}
+
+	arch_spin_unlock(&sb->lock);
+	return rc;
+}
+
 static int xive_native_validate_queue_size(u32 qshift)
 {
 	/*
@@ -670,6 +702,9 @@ static int kvmppc_xive_native_set_attr(struct kvm_device *dev,
 	case KVM_DEV_XIVE_GRP_EQ_CONFIG:
 		return kvmppc_xive_native_set_queue_config(xive, attr->attr,
 							   attr->addr);
+	case KVM_DEV_XIVE_GRP_SOURCE_SYNC:
+		return kvmppc_xive_native_sync_source(xive, attr->attr,
+						      attr->addr);
 	}
 	return -ENXIO;
 }
@@ -699,6 +734,7 @@ static int kvmppc_xive_native_has_attr(struct kvm_device *dev,
 		break;
 	case KVM_DEV_XIVE_GRP_SOURCE:
 	case KVM_DEV_XIVE_GRP_SOURCE_CONFIG:
+	case KVM_DEV_XIVE_GRP_SOURCE_SYNC:
 		if (attr->attr >= KVMPPC_XIVE_FIRST_IRQ &&
 		    attr->attr < KVMPPC_XIVE_NR_IRQS)
 			return 0;
Documentation/virtual/kvm/devices/xive.txt +8 -0
index 429cbc4cf960..1e7f19d7594b 100644
@@ -92,3 +92,11 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
     -EINVAL: Invalid queue address
     -EFAULT: Invalid user pointer for attr->addr.
     -EIO:    Configuration of the underlying HW failed
+
+  5. KVM_DEV_XIVE_GRP_SOURCE_SYNC (write only)
+  Synchronize the source to flush event notifications
+  Attributes:
+    Interrupt source number  (64-bit)
+  Errors:
+    -ENOENT: Unknown source number
+    -EINVAL: Not initialized source number
Signature · 1 line
2.20.1
#9 lore raw

PATCH v6 9/17 KVM: PPC: Book3S HV: XIVE: add a control to dirty the XIVE EQ pages

In reply to Cédric Le Goater

When migration of a VM is initiated, a first copy of the RAM is transferred to the destination before the VM is stopped, but there is no guarantee that the EQ pages in which the event notifications are queued have not been modified.

To make sure migration will capture a consistent memory state, the XIVE device should perform a XIVE quiesce sequence to stop the flow of event notifications and stabilize the EQs. This is the purpose of the KVM_DEV_XIVE_EQ_SYNC control which will also marks the EQ pages dirty to force their transfer.

---
 arch/powerpc/include/uapi/asm/kvm.h        |  1 +
 arch/powerpc/kvm/book3s_xive_native.c      | 85 ++++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt | 29 ++++++++
 3 files changed, 115 insertions(+)
arch/powerpc/include/uapi/asm/kvm.h +1 -0
index e4abe30f6fc6..12744608a61c 100644
@@ -680,6 +680,7 @@ struct kvm_ppc_cpu_char {
 /* POWER9 XIVE Native Interrupt Controller */
 #define KVM_DEV_XIVE_GRP_CTRL		1
 #define   KVM_DEV_XIVE_RESET		1
+#define   KVM_DEV_XIVE_EQ_SYNC		2
 #define KVM_DEV_XIVE_GRP_SOURCE		2	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_SOURCE_CONFIG	3	/* 64-bit source identifier */
 #define KVM_DEV_XIVE_GRP_EQ_CONFIG	4	/* 64-bit EQ identifier */
arch/powerpc/kvm/book3s_xive_native.c +85 -0

103 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +29 -0
index 1e7f19d7594b..7ffd4c7be7b5 100644
@@ -23,6 +23,12 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
     queues. To be used by kexec and kdump.
     Errors: none
 
+    1.2 KVM_DEV_XIVE_EQ_SYNC (write only)
+    Sync all the sources and queues and mark the EQ pages dirty. This
+    to make sure that a consistent memory state is captured when
+    migrating the VM.
+    Errors: none
+
   2. KVM_DEV_XIVE_GRP_SOURCE (write only)
   Initializes a new source in the XIVE device and mask it.
   Attributes:
@@ -100,3 +106,26 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
   Errors:
     -ENOENT: Unknown source number
     -EINVAL: Not initialized source number
+
+* Migration:
+
+  Saving the state of a VM using the XIVE native exploitation mode
+  should follow a specific sequence. When the VM is stopped :
+
+  1. Mask all sources (PQ) to stop the flow of events.
+
+  2. Sync the XIVE device with the KVM control KVM_DEV_XIVE_EQ_SYNC to
+  flush any in-flight event notification and to stabilize the EQs. At
+  this stage, the EQ pages are marked dirty to make sure they are
+  transferred in the migration sequence.
+
+  3. Capture the state of the source targeting, the EQs configuration
+  and the state of thread interrupt context registers.
+
+  Restore is similar :
+
+  1. Restore the EQ configuration. As targeting depends on it.
+  2. Restore targeting
+  3. Restore the thread interrupt contexts
+  4. Restore the source states
+  5. Let the vCPU run
Signature · 1 line
2.20.1
#10 lore raw

PATCH v6 10/17 KVM: PPC: Book3S HV: XIVE: add get/set accessors for the VP XIVE state

In reply to Cédric Le Goater

The state of the thread interrupt management registers needs to be collected for migration. These registers are cached under the 'xive_saved_state.w01' field of the VCPU when the VPCU context is pulled from the HW thread. An OPAL call retrieves the backup of the IPB register in the underlying XIVE NVT structure and merges it in the KVM state.

---
 arch/powerpc/include/asm/kvm_ppc.h         | 11 ++++
 arch/powerpc/include/uapi/asm/kvm.h        |  2 +
 arch/powerpc/kvm/book3s.c                  | 24 +++++++
 arch/powerpc/kvm/book3s_xive_native.c      | 76 ++++++++++++++++++++++
 Documentation/virtual/kvm/api.txt          |  1 +
 Documentation/virtual/kvm/devices/xive.txt | 17 +++++
 6 files changed, 131 insertions(+)
arch/powerpc/include/asm/kvm_ppc.h +11 -0
index 6928a35ac3c7..0579c9b253db 100644
@@ -273,6 +273,7 @@ union kvmppc_one_reg {
 		u64	addr;
 		u64	length;
 	}	vpaval;
+	u64	xive_timaval[2];
 };
 
 struct kvmppc_ops {
@@ -605,6 +606,10 @@ extern int kvmppc_xive_native_connect_vcpu(struct kvm_device *dev,
 extern void kvmppc_xive_native_cleanup_vcpu(struct kvm_vcpu *vcpu);
 extern void kvmppc_xive_native_init_module(void);
 extern void kvmppc_xive_native_exit_module(void);
+extern int kvmppc_xive_native_get_vp(struct kvm_vcpu *vcpu,
+				     union kvmppc_one_reg *val);
+extern int kvmppc_xive_native_set_vp(struct kvm_vcpu *vcpu,
+				     union kvmppc_one_reg *val);
 
 #else
 static inline int kvmppc_xive_set_xive(struct kvm *kvm, u32 irq, u32 server,
@@ -637,6 +642,12 @@ static inline int kvmppc_xive_native_connect_vcpu(struct kvm_device *dev,
 static inline void kvmppc_xive_native_cleanup_vcpu(struct kvm_vcpu *vcpu) { }
 static inline void kvmppc_xive_native_init_module(void) { }
 static inline void kvmppc_xive_native_exit_module(void) { }
+static inline int kvmppc_xive_native_get_vp(struct kvm_vcpu *vcpu,
+					    union kvmppc_one_reg *val)
+{ return 0; }
+static inline int kvmppc_xive_native_set_vp(struct kvm_vcpu *vcpu,
+					    union kvmppc_one_reg *val)
+{ return -ENOENT; }
 
 #endif /* CONFIG_KVM_XIVE */
 
arch/powerpc/include/uapi/asm/kvm.h +2 -0
index 12744608a61c..cd3f16b70a2e 100644
@@ -482,6 +482,8 @@ struct kvm_ppc_cpu_char {
 #define  KVM_REG_PPC_ICP_PPRI_SHIFT	16	/* pending irq priority */
 #define  KVM_REG_PPC_ICP_PPRI_MASK	0xff
 
+#define KVM_REG_PPC_VP_STATE	(KVM_REG_PPC | KVM_REG_SIZE_U128 | 0x8d)
+
 /* Device control API: PPC-specific devices */
 #define KVM_DEV_MPIC_GRP_MISC		1
 #define   KVM_DEV_MPIC_BASE_ADDR	0	/* 64-bit */
arch/powerpc/kvm/book3s.c +24 -0
index 7c3348fa27e1..efd15101eef0 100644
@@ -651,6 +651,18 @@ int kvmppc_get_one_reg(struct kvm_vcpu *vcpu, u64 id,
 				*val = get_reg_val(id, kvmppc_xics_get_icp(vcpu));
 			break;
 #endif /* CONFIG_KVM_XICS */
+#ifdef CONFIG_KVM_XIVE
+		case KVM_REG_PPC_VP_STATE:
+			if (!vcpu->arch.xive_vcpu) {
+				r = -ENXIO;
+				break;
+			}
+			if (xive_enabled())
+				r = kvmppc_xive_native_get_vp(vcpu, val);
+			else
+				r = -ENXIO;
+			break;
+#endif /* CONFIG_KVM_XIVE */
 		case KVM_REG_PPC_FSCR:
 			*val = get_reg_val(id, vcpu->arch.fscr);
 			break;
@@ -724,6 +736,18 @@ int kvmppc_set_one_reg(struct kvm_vcpu *vcpu, u64 id,
 				r = kvmppc_xics_set_icp(vcpu, set_reg_val(id, *val));
 			break;
 #endif /* CONFIG_KVM_XICS */
+#ifdef CONFIG_KVM_XIVE
+		case KVM_REG_PPC_VP_STATE:
+			if (!vcpu->arch.xive_vcpu) {
+				r = -ENXIO;
+				break;
+			}
+			if (xive_enabled())
+				r = kvmppc_xive_native_set_vp(vcpu, val);
+			else
+				r = -ENXIO;
+			break;
+#endif /* CONFIG_KVM_XIVE */
 		case KVM_REG_PPC_FSCR:
 			vcpu->arch.fscr = set_reg_val(id, *val);
 			break;
arch/powerpc/kvm/book3s_xive_native.c +76 -0

82 lines not on this page · raw message

Documentation/virtual/kvm/api.txt +1 -0
index e38eb17b7be6..5b505520a616 100644
@@ -1985,6 +1985,7 @@ registers, find a list below:
   PPC   | KVM_REG_PPC_TLB3PS            | 32
   PPC   | KVM_REG_PPC_EPTCFG            | 32
   PPC   | KVM_REG_PPC_ICP_STATE         | 64
+  PPC   | KVM_REG_PPC_VP_STATE          | 128
   PPC   | KVM_REG_PPC_TB_OFFSET         | 64
   PPC   | KVM_REG_PPC_SPMC1             | 32
   PPC   | KVM_REG_PPC_SPMC2             | 32
Documentation/virtual/kvm/devices/xive.txt +17 -0
index 7ffd4c7be7b5..525d1eebcf34 100644
@@ -107,6 +107,23 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
     -ENOENT: Unknown source number
     -EINVAL: Not initialized source number
 
+* VCPU state
+
+  The XIVE IC maintains VP interrupt state in an internal structure
+  called the NVT. When a VP is not dispatched on a HW processor
+  thread, this structure can be updated by HW if the VP is the target
+  of an event notification.
+
+  It is important for migration to capture the cached IPB from the NVT
+  as it synthesizes the priorities of the pending interrupts. We
+  capture a bit more to report debug information.
+
+  KVM_REG_PPC_VP_STATE (2 * 64bits)
+  bits:     |  63  ....  32  |  31  ....  0  |
+  values:   |   TIMA word0   |   TIMA word1  |
+  bits:     | 127       ..........       64  |
+  values:   |            unused              |
+
 * Migration:
 
   Saving the state of a VM using the XIVE native exploitation mode
Signature · 1 line
2.20.1
#11 lore raw

PATCH v6 11/17 KVM: introduce a 'mmap' method for KVM devices

In reply to Cédric Le Goater

Some KVM devices will want to handle special mappings related to the underlying HW. For instance, the XIVE interrupt controller of the POWER9 processor has MMIO pages for thread interrupt management and for interrupt source control that need to be exposed to the guest when the OS has the required support.

---
 include/linux/kvm_host.h |  1 +
 virt/kvm/kvm_main.c      | 11 +++++++++++
 2 files changed, 12 insertions(+)
include/linux/kvm_host.h +1 -0
index 9d55c63db09b..831d963451d8 100644
@@ -1245,6 +1245,7 @@ struct kvm_device_ops {
 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
 	long (*ioctl)(struct kvm_device *dev, unsigned int ioctl,
 		      unsigned long arg);
+	int (*mmap)(struct kvm_device *dev, struct vm_area_struct *vma);
 };
 
 void kvm_device_get(struct kvm_device *dev);
virt/kvm/kvm_main.c +11 -0
index 55fe8e20d8fd..ea2018ae1cd7 100644
@@ -2884,6 +2884,16 @@ static long kvm_vcpu_compat_ioctl(struct file *filp,
 }
 #endif
 
+static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+	struct kvm_device *dev = filp->private_data;
+
+	if (dev->ops->mmap)
+		return dev->ops->mmap(dev, vma);
+
+	return -ENODEV;
+}
+
 static int kvm_device_ioctl_attr(struct kvm_device *dev,
 				 int (*accessor)(struct kvm_device *dev,
 						 struct kvm_device_attr *attr),
@@ -2936,6 +2946,7 @@ static const struct file_operations kvm_device_fops = {
 	.unlocked_ioctl = kvm_device_ioctl,
 	.release = kvm_device_release,
 	KVM_COMPAT(kvm_device_ioctl),
+	.mmap = kvm_device_mmap,
 };
 
 struct kvm_device *kvm_device_from_filp(struct file *filp)
Signature · 1 line
2.20.1
#12 lore raw

PATCH v6 12/17 KVM: PPC: Book3S HV: XIVE: add a TIMA mapping

In reply to Cédric Le Goater

Each thread has an associated Thread Interrupt Management context composed of a set of registers. These registers let the thread handle priority management and interrupt acknowledgment. The most important are :

- Interrupt Pending Buffer (IPB)
- Current Processor Priority (CPPR)
- Notification Source Register (NSR)

They are exposed to software in four different pages each proposing a view with a different privilege. The first page is for the physical thread context and the second for the hypervisor. Only the third (operating system) and the fourth (user level) are exposed the guest.

A custom VM fault handler will populate the VMA with the appropriate pages, which should only be the OS page for now.

---
 arch/powerpc/include/asm/xive.h            |  1 +
 arch/powerpc/include/uapi/asm/kvm.h        |  2 ++
 arch/powerpc/kvm/book3s_xive_native.c      | 39 ++++++++++++++++++++++
 arch/powerpc/sysdev/xive/native.c          | 11 ++++++
 Documentation/virtual/kvm/devices/xive.txt | 23 +++++++++++++
 5 files changed, 76 insertions(+)
arch/powerpc/include/asm/xive.h +1 -0
index c4e88abd3b67..eaf76f57023a 100644
@@ -23,6 +23,7 @@
  * same offset regardless of where the code is executing
  */
 extern void __iomem *xive_tima;
+extern unsigned long xive_tima_os;
 
 /*
  * Offset in the TM area of our current execution level (provided by
arch/powerpc/include/uapi/asm/kvm.h +2 -0
index cd3f16b70a2e..0998e8edc91a 100644
@@ -720,4 +720,6 @@ struct kvm_ppc_xive_eq {
 
 #define KVM_XIVE_EQ_ALWAYS_NOTIFY	0x00000001
 
+#define KVM_XIVE_TIMA_PAGE_OFFSET	0
+
 #endif /* __LINUX_KVM_POWERPC_H */
arch/powerpc/kvm/book3s_xive_native.c +39 -0
index 2f9d5e9439a6..f5314da0cb45 100644
@@ -165,6 +165,44 @@ int kvmppc_xive_native_connect_vcpu(struct kvm_device *dev,
 	return rc;
 }
 
+static vm_fault_t xive_native_tima_fault(struct vm_fault *vmf)
+{
+	struct vm_area_struct *vma = vmf->vma;
+
+	switch (vmf->pgoff - vma->vm_pgoff) {
+	case 0: /* HW - forbid access */
+	case 1: /* HV - forbid access */
+		return VM_FAULT_SIGBUS;
+	case 2: /* OS */
+		vmf_insert_pfn(vma, vmf->address, xive_tima_os >> PAGE_SHIFT);
+		return VM_FAULT_NOPAGE;
+	case 3: /* USER - TODO */
+	default:
+		return VM_FAULT_SIGBUS;
+	}
+}
+
+static const struct vm_operations_struct xive_native_tima_vmops = {
+	.fault = xive_native_tima_fault,
+};
+
+static int kvmppc_xive_native_mmap(struct kvm_device *dev,
+				   struct vm_area_struct *vma)
+{
+	/* We only allow mappings at fixed offset for now */
+	if (vma->vm_pgoff = KVM_XIVE_TIMA_PAGE_OFFSET) {
+		if (vma_pages(vma) > 4)
+			return -EINVAL;
+		vma->vm_ops = &xive_native_tima_vmops;
+	} else {
+		return -EINVAL;
+	}
+
+	vma->vm_flags |= VM_IO | VM_PFNMAP;
+	vma->vm_page_prot = pgprot_noncached_wc(vma->vm_page_prot);
+	return 0;
+}
+
 static int kvmppc_xive_native_set_source(struct kvmppc_xive *xive, long irq,
 					 u64 addr)
 {
@@ -1050,6 +1088,7 @@ struct kvm_device_ops kvm_xive_native_ops = {
 	.set_attr = kvmppc_xive_native_set_attr,
 	.get_attr = kvmppc_xive_native_get_attr,
 	.has_attr = kvmppc_xive_native_has_attr,
+	.mmap = kvmppc_xive_native_mmap,
 };
 
 void kvmppc_xive_native_init_module(void)
arch/powerpc/sysdev/xive/native.c +11 -0
index 0c037e933e55..7782201e5fe8 100644
@@ -521,6 +521,9 @@ u32 xive_native_default_eq_shift(void)
 }
 EXPORT_SYMBOL_GPL(xive_native_default_eq_shift);
 
+unsigned long xive_tima_os;
+EXPORT_SYMBOL_GPL(xive_tima_os);
+
 bool __init xive_native_init(void)
 {
 	struct device_node *np;
@@ -573,6 +576,14 @@ bool __init xive_native_init(void)
 	for_each_possible_cpu(cpu)
 		kvmppc_set_xive_tima(cpu, r.start, tima);
 
+	/* Resource 2 is OS window */
+	if (of_address_to_resource(np, 2, &r)) {
+		pr_err("Failed to get thread mgmnt area resource\n");
+		return false;
+	}
+
+	xive_tima_os = r.start;
+
 	/* Grab size of provisionning pages */
 	xive_parse_provisioning(np);
 
Documentation/virtual/kvm/devices/xive.txt +23 -0
index 525d1eebcf34..0cd7847ec38a 100644
@@ -13,6 +13,29 @@ requires a POWER9 host and the guest OS should have support for the
 XIVE native exploitation interrupt mode. If not, it should run using
 the legacy interrupt mode, referred as XICS (POWER7/8).
 
+* Device Mappings
+
+  The KVM device exposes different MMIO ranges of the XIVE HW which
+  are required for interrupt management. These are exposed to the
+  guest in VMAs populated with a custom VM fault handler.
+
+  1. Thread Interrupt Management Area (TIMA)
+
+  Each thread has an associated Thread Interrupt Management context
+  composed of a set of registers. These registers let the thread
+  handle priority management and interrupt acknowledgment. The most
+  important are :
+
+      - Interrupt Pending Buffer     (IPB)
+      - Current Processor Priority   (CPPR)
+      - Notification Source Register (NSR)
+
+  They are exposed to software in four different pages each proposing
+  a view with a different privilege. The first page is for the
+  physical thread context and the second for the hypervisor. Only the
+  third (operating system) and the fourth (user level) are exposed the
+  guest.
+
 * Groups:
 
   1. KVM_DEV_XIVE_GRP_CTRL
Signature · 1 line
2.20.1
#13 lore raw

PATCH v6 13/17 KVM: PPC: Book3S HV: XIVE: add a mapping for the source ESB pages

In reply to Cédric Le Goater

Each source is associated with an Event State Buffer (ESB) with a even/odd pair of pages which provides commands to manage the source:
to trigger, to EOI, to turn off the source for instance.

The custom VM fault handler will deduce the guest IRQ number from the offset of the fault, and the ESB page of the associated XIVE interrupt will be inserted into the VMA using the internal structure caching information on the interrupts.

---
 arch/powerpc/include/uapi/asm/kvm.h        |  1 +
 arch/powerpc/kvm/book3s_xive_native.c      | 57 ++++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt |  7 +++
 3 files changed, 65 insertions(+)
arch/powerpc/include/uapi/asm/kvm.h +1 -0
index 0998e8edc91a..b0f72dea8b11 100644
@@ -721,5 +721,6 @@ struct kvm_ppc_xive_eq {
 #define KVM_XIVE_EQ_ALWAYS_NOTIFY	0x00000001
 
 #define KVM_XIVE_TIMA_PAGE_OFFSET	0
+#define KVM_XIVE_ESB_PAGE_OFFSET	4
 
 #endif /* __LINUX_KVM_POWERPC_H */
arch/powerpc/kvm/book3s_xive_native.c +57 -0

69 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +7 -0
index 0cd7847ec38a..69ee62d3d4dc 100644
@@ -36,6 +36,13 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
   third (operating system) and the fourth (user level) are exposed the
   guest.
 
+  2. Event State Buffer (ESB)
+
+  Each source is associated with an Event State Buffer (ESB) with
+  either a pair of even/odd pair of pages which provides commands to
+  manage the source: to trigger, to EOI, to turn off the source for
+  instance.
+
 * Groups:
 
   1. KVM_DEV_XIVE_GRP_CTRL
Signature · 1 line
2.20.1
#14 lore raw

PATCH v6 14/17 KVM: PPC: Book3S HV: XIVE: add passthrough support

In reply to Cédric Le Goater

The KVM XICS-over-XIVE device and the proposed KVM XIVE native device implement an IRQ space for the guest using the generic IPI interrupts of the XIVE IC controller. These interrupts are allocated at the OPAL level and "mapped" into the guest IRQ number space in the range 0-0x1FFF. Interrupt management is performed in the XIVE way: using loads and stores on the addresses of the XIVE IPI interrupt ESB pages.

Both KVM devices share the same internal structure caching information on the interrupts, among which the xive_irq_data struct containing the addresses of the IPI ESB pages and an extra one in case of pass-through. The later contains the addresses of the ESB pages of the underlying HW controller interrupts, PHB4 in all cases for now.

A guest, when running in the XICS legacy interrupt mode, lets the KVM XICS-over-XIVE device "handle" interrupt management, that is to perform the loads and stores on the addresses of the ESB pages of the guest interrupts. However, when running in XIVE native exploitation mode, the KVM XIVE native device exposes the interrupt ESB pages to the guest and lets the guest perform directly the loads and stores.

The VMA exposing the ESB pages make use of a custom VM fault handler which role is to populate the VMA with appropriate pages. When a fault occurs, the guest IRQ number is deduced from the offset, and the ESB pages of associated XIVE IPI interrupt are inserted in the VMA (using the internal structure caching information on the interrupts).

Supporting device passthrough in the guest running in XIVE native exploitation mode adds some extra refinements because the ESB pages of a different HW controller (PHB4) need to be exposed to the guest along with the initial IPI ESB pages of the XIVE IC controller. But the overall mechanic is the same.

When the device HW irqs are mapped into or unmapped from the guest IRQ number space, the passthru_irq helpers, kvmppc_xive_set_mapped() and kvmppc_xive_clr_mapped(), are called to record or clear the passthrough interrupt information and to perform the switch.

The approach taken by this patch is to clear the ESB pages of the guest IRQ number being mapped and let the VM fault handler repopulate. The handler will insert the ESB page corresponding to the HW interrupt of the device being passed-through or the initial IPI ESB page if the device is being removed.

---
 arch/powerpc/kvm/book3s_xive.h             |  9 +++++
 arch/powerpc/kvm/book3s_xive.c             | 15 ++++++++
 arch/powerpc/kvm/book3s_xive_native.c      | 41 ++++++++++++++++++++++
 Documentation/virtual/kvm/devices/xive.txt | 19 ++++++++++
 4 files changed, 84 insertions(+)
arch/powerpc/kvm/book3s_xive.h +9 -0
index 622f594d93e1..e011622dc038 100644
@@ -94,6 +94,11 @@ struct kvmppc_xive_src_block {
 	struct kvmppc_xive_irq_state irq_state[KVMPPC_XICS_IRQ_PER_ICS];
 };
 
+struct kvmppc_xive;
+
+struct kvmppc_xive_ops {
+	int (*reset_mapped)(struct kvm *kvm, unsigned long guest_irq);
+};
 
 struct kvmppc_xive {
 	struct kvm *kvm;
@@ -132,6 +137,10 @@ struct kvmppc_xive {
 
 	/* Flags */
 	u8	single_escalation;
+
+	struct kvmppc_xive_ops *ops;
+	struct address_space   *mapping;
+	struct mutex mapping_lock;
 };
 
 #define KVMPPC_XIVE_Q_COUNT	8
arch/powerpc/kvm/book3s_xive.c +15 -0
index c1b7aa7dbc28..480a3fc6b9fd 100644
@@ -937,6 +937,13 @@ int kvmppc_xive_set_mapped(struct kvm *kvm, unsigned long guest_irq,
 	/* Turn the IPI hard off */
 	xive_vm_esb_load(&state->ipi_data, XIVE_ESB_SET_PQ_01);
 
+	/*
+	 * Reset ESB guest mapping. Needed when ESB pages are exposed
+	 * to the guest in XIVE native mode
+	 */
+	if (xive->ops && xive->ops->reset_mapped)
+		xive->ops->reset_mapped(kvm, guest_irq);
+
 	/* Grab info about irq */
 	state->pt_number = hw_irq;
 	state->pt_data = irq_data_get_irq_handler_data(host_data);
@@ -1022,6 +1029,14 @@ int kvmppc_xive_clr_mapped(struct kvm *kvm, unsigned long guest_irq,
 	state->pt_number = 0;
 	state->pt_data = NULL;
 
+	/*
+	 * Reset ESB guest mapping. Needed when ESB pages are exposed
+	 * to the guest in XIVE native mode
+	 */
+	if (xive->ops && xive->ops->reset_mapped) {
+		xive->ops->reset_mapped(kvm, guest_irq);
+	}
+
 	/* Reconfigure the IPI */
 	xive_native_configure_irq(state->ipi_number,
 				  kvmppc_xive_vp(xive, state->act_server),
arch/powerpc/kvm/book3s_xive_native.c +41 -0

77 lines not on this page · raw message

Documentation/virtual/kvm/devices/xive.txt +19 -0
index 69ee62d3d4dc..9a24a4525253 100644
@@ -43,6 +43,25 @@ the legacy interrupt mode, referred as XICS (POWER7/8).
   manage the source: to trigger, to EOI, to turn off the source for
   instance.
 
+  3. Device pass-through
+
+  When a device is passed-through into the guest, the source
+  interrupts are from a different HW controller (PHB4) and the ESB
+  pages exposed to the guest should accommadate this change.
+
+  The passthru_irq helpers, kvmppc_xive_set_mapped() and
+  kvmppc_xive_clr_mapped() are called when the device HW irqs are
+  mapped into or unmapped from the guest IRQ number space. The KVM
+  device extends these helpers to clear the ESB pages of the guest IRQ
+  number being mapped and then lets the VM fault handler repopulate.
+  The handler will insert the ESB page corresponding to the HW
+  interrupt of the device being passed-through or the initial IPI ESB
+  page if the device has being removed.
+
+  The ESB remapping is fully transparent to the guest and the OS
+  device driver. All handling is done within VFIO and the above
+  helpers in KVM-PPC.
+
 * Groups:
 
   1. KVM_DEV_XIVE_GRP_CTRL
Signature · 1 line
2.20.1
#15 lore raw

Re: [PATCH v6 14/17] KVM: PPC: Book3S HV: XIVE: add passthrough support

Quoted 37 lines · Cédric Le Goater

The KVM XICS-over-XIVE device and the proposed KVM XIVE native device implement an IRQ space for the guest using the generic IPI interrupts of the XIVE IC controller. These interrupts are allocated at the OPAL level and "mapped" into the guest IRQ number space in the range 0-0x1FFF. Interrupt management is performed in the XIVE way: using loads and stores on the addresses of the XIVE IPI interrupt ESB pages.

Both KVM devices share the same internal structure caching information on the interrupts, among which the xive_irq_data struct containing the addresses of the IPI ESB pages and an extra one in case of pass-through. The later contains the addresses of the ESB pages of the underlying HW controller interrupts, PHB4 in all cases for now.

A guest, when running in the XICS legacy interrupt mode, lets the KVM XICS-over-XIVE device "handle" interrupt management, that is to perform the loads and stores on the addresses of the ESB pages of the guest interrupts. However, when running in XIVE native exploitation mode, the KVM XIVE native device exposes the interrupt ESB pages to the guest and lets the guest perform directly the loads and stores.

The VMA exposing the ESB pages make use of a custom VM fault handler which role is to populate the VMA with appropriate pages. When a fault occurs, the guest IRQ number is deduced from the offset, and the ESB pages of associated XIVE IPI interrupt are inserted in the VMA (using the internal structure caching information on the interrupts).

Supporting device passthrough in the guest running in XIVE native exploitation mode adds some extra refinements because the ESB pages of a different HW controller (PHB4) need to be exposed to the guest along with the initial IPI ESB pages of the XIVE IC controller. But the overall mechanic is the same.

When the device HW irqs are mapped into or unmapped from the guest IRQ number space, the passthru_irq helpers, kvmppc_xive_set_mapped() and kvmppc_xive_clr_mapped(), are called to record or clear the passthrough interrupt information and to perform the switch.

Quoted 5 lines

The approach taken by this patch is to clear the ESB pages of the guest IRQ number being mapped and let the VM fault handler repopulate. The handler will insert the ESB page corresponding to the HW interrupt of the device being passed-through or the initial IPI ESB page if the device is being removed.

One comment below:

arch/powerpc/kvm/book3s_xive_native.c static int kvmppc_xive_native_mmap(struct kvm_device *dev, · line 298

Quoted 11 lines
arch/powerpc/kvm/book3s_xive_native.c +7 -0
@@ -257,6 +289,13 @@ static int kvmppc_xive_native_mmap(struct kvm_device *dev,
 
 	vma->vm_flags |= VM_IO | VM_PFNMAP;
 	vma->vm_page_prot = pgprot_noncached_wc(vma->vm_page_prot);
+
+	/*
+	 * Grab the KVM device file address_space to be able to clear
+	 * the ESB pages mapping when a device is passed-through into
+	 * the guest.
+	 */
+	xive->mapping = vma->vm_file->f_mapping;

I'm worried about the lifetime of this pointer. At the least you should clear it in the release function for the device, when you get one. It looks like you should hold the xive->mapping_lock around the clearing. I think that should be OK then since the release function won't get called until all of the mmaps of the fd have been destroyed, as I understand things.

Paul.

#16 lore raw

Re: [PATCH v6 14/17] KVM: PPC: Book3S HV: XIVE: add passthrough support

Quoted 48 lines · Paul Mackerras
Quoted 37 lines · Cédric Le Goater

The KVM XICS-over-XIVE device and the proposed KVM XIVE native device implement an IRQ space for the guest using the generic IPI interrupts of the XIVE IC controller. These interrupts are allocated at the OPAL level and "mapped" into the guest IRQ number space in the range 0-0x1FFF. Interrupt management is performed in the XIVE way: using loads and stores on the addresses of the XIVE IPI interrupt ESB pages.

Both KVM devices share the same internal structure caching information on the interrupts, among which the xive_irq_data struct containing the addresses of the IPI ESB pages and an extra one in case of pass-through. The later contains the addresses of the ESB pages of the underlying HW controller interrupts, PHB4 in all cases for now.

A guest, when running in the XICS legacy interrupt mode, lets the KVM XICS-over-XIVE device "handle" interrupt management, that is to perform the loads and stores on the addresses of the ESB pages of the guest interrupts. However, when running in XIVE native exploitation mode, the KVM XIVE native device exposes the interrupt ESB pages to the guest and lets the guest perform directly the loads and stores.

The VMA exposing the ESB pages make use of a custom VM fault handler which role is to populate the VMA with appropriate pages. When a fault occurs, the guest IRQ number is deduced from the offset, and the ESB pages of associated XIVE IPI interrupt are inserted in the VMA (using the internal structure caching information on the interrupts).

Supporting device passthrough in the guest running in XIVE native exploitation mode adds some extra refinements because the ESB pages of a different HW controller (PHB4) need to be exposed to the guest along with the initial IPI ESB pages of the XIVE IC controller. But the overall mechanic is the same.

When the device HW irqs are mapped into or unmapped from the guest IRQ number space, the passthru_irq helpers, kvmppc_xive_set_mapped() and kvmppc_xive_clr_mapped(), are called to record or clear the passthrough interrupt information and to perform the switch.

Quoted 5 lines

The approach taken by this patch is to clear the ESB pages of the guest IRQ number being mapped and let the VM fault handler repopulate. The handler will insert the ESB page corresponding to the HW interrupt of the device being passed-through or the initial IPI ESB page if the device is being removed.

One comment below:

Quoted 2 lines
(quoted diff) +0 -0
@@ -257,6 +289,13 @@ static int kvmppc_xive_native_mmap(struct kvm_device *dev,
 
Quoted 14 lines

arch/powerpc/kvm/book3s_xive_native.c static int kvmppc_xive_native_mmap(struct kvm_device *dev, · line 298

Quoted 9 lines
arch/powerpc/kvm/book3s_xive_native.c +7 -0
 	vma->vm_flags |= VM_IO | VM_PFNMAP;
 	vma->vm_page_prot = pgprot_noncached_wc(vma->vm_page_prot);
+
+	/*
+	 * Grab the KVM device file address_space to be able to clear
+	 * the ESB pages mapping when a device is passed-through into
+	 * the guest.
+	 */
+	xive->mapping = vma->vm_file->f_mapping;

I'm worried about the lifetime of this pointer. At the least you should clear it in the release function for the device, when you get one. It looks like you should hold the xive->mapping_lock around the clearing.

Yes. This is should be cleaner.

Quoted 3 lines

I think that should be OK then since the release function won't get called until all of the mmaps of the fd have been destroyed, as I understand things.

and you also have to close the fd by exiting QEMU or resetting the guest.

Anyhow, I will send a cleanup in a followup patch.

Thanks,

C.

#17 lore raw

PATCH v6 15/17 KVM: PPC: Book3S HV: XIVE: activate XIVE exploitation mode

In reply to Cédric Le Goater

Full support for the XIVE native exploitation mode is now available, advertise the capability KVM_CAP_PPC_IRQ_XIVE for guests running on PowerNV KVM Hypervisors only. Support for nested guests (pseries KVM Hypervisor) is not yet available. XIVE should also have been activated which is default setting on POWER9 systems running a recent Linux kernel.

---
 arch/powerpc/kvm/powerpc.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
arch/powerpc/kvm/powerpc.c +4 -3
index b0858ee61460..f54926c78320 100644
@@ -573,10 +573,11 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 #ifdef CONFIG_KVM_XIVE
 	case KVM_CAP_PPC_IRQ_XIVE:
 		/*
-		 * Return false until all the XIVE infrastructure is
-		 * in place including support for migration.
+		 * We need XIVE to be enabled on the platform (implies
+		 * a POWER9 processor) and the PowerNV platform, as
+		 * nested is not yet supported.
 		 */
-		r = 0;
+		r = xive_enabled() && !!cpu_has_feature(CPU_FTR_HVMODE);
 		break;
 #endif
 
Signature · 1 line
2.20.1
#18 lore raw

PATCH v6 16/17 KVM: introduce a 'release' method for KVM devices

In reply to Cédric Le Goater

When a P9 sPAPR VM boots, the CAS negotiation process determines which interrupt mode to use (XICS legacy or XIVE native) and invokes a machine reset to activate the chosen mode.

To be able to switch from one interrupt mode to another, we introduce the capability to release a KVM device without destroying the VM. The KVM device interface is extended with a new 'release' method which is called when the file descriptor of the device is closed.

Once 'release' is called, the 'destroy' method will not be called anymore as the device is removed from the device list of the VM.

---
 include/linux/kvm_host.h |  9 +++++++++
 virt/kvm/kvm_main.c      | 13 +++++++++++++
 2 files changed, 22 insertions(+)
include/linux/kvm_host.h +9 -0
index 831d963451d8..722692e2f745 100644
@@ -1240,6 +1240,15 @@ struct kvm_device_ops {
 	 */
 	void (*destroy)(struct kvm_device *dev);
 
+	/*
+	 * Release is an alternative method to free the device. It is
+	 * called when the device file descriptor is closed. Once
+	 * release is called, the destroy method will not be called
+	 * anymore as the device is removed from the device list of
+	 * the VM. kvm->lock is held.
+	 */
+	void (*release)(struct kvm_device *dev);
+
 	int (*set_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
 	int (*get_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
 	int (*has_attr)(struct kvm_device *dev, struct kvm_device_attr *attr);
virt/kvm/kvm_main.c +13 -0
index ea2018ae1cd7..ea2619d5ca98 100644
@@ -2938,6 +2938,19 @@ static int kvm_device_release(struct inode *inode, struct file *filp)
 	struct kvm_device *dev = filp->private_data;
 	struct kvm *kvm = dev->kvm;
 
+	if (!dev)
+		return -ENODEV;
+
+	if (dev->kvm != kvm)
+		return -EPERM;
+
+	if (dev->ops->release) {
+		mutex_lock(&kvm->lock);
+		list_del(&dev->vm_node);
+		dev->ops->release(dev);
+		mutex_unlock(&kvm->lock);
+	}
+
 	kvm_put_kvm(kvm);
 	return 0;
 }
Signature · 1 line
2.20.1
#19 lore raw

PATCH v6 17/17 KVM: PPC: Book3S HV: XIVE: replace the 'destroy' method by a 'release' method

In reply to Cédric Le Goater

When a P9 sPAPR VM boots, the CAS negotiation process determines which interrupt mode to use (XICS legacy or XIVE native) and invokes a machine reset to activate the chosen mode.

We introduce 'release' methods for the XICS-on-XIVE and the XIVE native KVM devices which are called when the file descriptor of the device is closed after the TIMA and ESB pages have been unmapped. They perform the necessary cleanups : clear the vCPU interrupt presenters that could be attached and then destroy the device. The 'release' methods replace the 'destroy' methods as 'destroy' is not called anymore once 'release' is. Compatibility with older QEMU is nevertheless maintained.

This is not considered as a safe operation as the vCPUs are still running and could be referencing the KVM device through their presenters. To protect the system from any breakage, the kvmppc_xive objects representing both KVM devices are now stored in an array under the VM. Allocation is performed on first usage and memory is freed only when the VM exits.

---

Changes since v5 :

- replaced the kvmppc_xive device array with named kvmppc_xive structs.
- in kvmppc_xive_cleanup_vcpu(), replaced xc->xive by vcpu->kvm->arch.xive which is always valid in this context
- in kvmppc_xive_release(), removed the comment on the race. The device being destroyed when the fd is closed, dev->private should always be valid when used. Also removed the kick_all_cpus_sync() to flush the IPIs.
- in the both release operation, removed the test on &kvm->online_vcpus.
- devices are now freed when the machine exits.

 arch/powerpc/include/asm/kvm_host.h   |  6 ++-
 arch/powerpc/kvm/book3s_xive.h        |  1 +
 arch/powerpc/kvm/book3s_xive.c        | 67 ++++++++++++++++++++++++---
 arch/powerpc/kvm/book3s_xive_native.c | 28 +++++++++--
 arch/powerpc/kvm/powerpc.c            |  9 ++++
 5 files changed, 99 insertions(+), 12 deletions(-)
arch/powerpc/include/asm/kvm_host.h +5 -1
index 9cc6abdce1b9..c7abc7c3c6cd 100644
@@ -313,7 +313,11 @@ struct kvm_arch {
 #endif
 #ifdef CONFIG_KVM_XICS
 	struct kvmppc_xics *xics;
-	struct kvmppc_xive *xive;
+	struct kvmppc_xive *xive;    /* Current XIVE device in use */
+	struct {
+		struct kvmppc_xive *native;
+		struct kvmppc_xive *xics_on_xive;
+	} xive_devices;
 	struct kvmppc_passthru_irqmap *pimap;
 #endif
 	struct kvmppc_ops *kvm_ops;
arch/powerpc/kvm/book3s_xive.h +1 -0
index e011622dc038..426146332984 100644
@@ -283,6 +283,7 @@ void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb);
 int kvmppc_xive_select_target(struct kvm *kvm, u32 *server, u8 prio);
 int kvmppc_xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio,
 				  bool single_escalation);
+struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type);
 
 #endif /* CONFIG_KVM_XICS */
 #endif /* _KVM_PPC_BOOK3S_XICS_H */
arch/powerpc/kvm/book3s_xive.c +61 -6

120 lines not on this page · raw message

arch/powerpc/kvm/book3s_xive_native.c +23 -5
index 62648f833adf..0497272a72fa 100644
@@ -964,15 +964,27 @@ static int kvmppc_xive_native_has_attr(struct kvm_device *dev,
 	return -ENXIO;
 }
 
-static void kvmppc_xive_native_free(struct kvm_device *dev)
+/*
+ * Called when device fd is closed
+ */
+static void kvmppc_xive_native_release(struct kvm_device *dev)
 {
 	struct kvmppc_xive *xive = dev->private;
 	struct kvm *kvm = xive->kvm;
+	struct kvm_vcpu *vcpu;
 	int i;
 
 	debugfs_remove(xive->dentry);
 
-	pr_devel("Destroying xive native device\n");
+	pr_devel("Releasing xive native device\n");
+
+	/*
+	 * When releasing the KVM device fd, the vCPUs can still be
+	 * running and we should clean up the vCPU interrupt
+	 * presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm)
+		kvmppc_xive_native_cleanup_vcpu(vcpu);
 
 	if (kvm)
 		kvm->arch.xive = NULL;
@@ -987,7 +999,13 @@ static void kvmppc_xive_native_free(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
-	kfree(xive);
+	/*
+	 * A reference of the kvmppc_xive pointer is now kept under
+	 * the xive_devices struct of the machine for reuse. It is
+	 * freed when the VM is destroyed for now until we fix all the
+	 * execution paths.
+	 */
+
 	kfree(dev);
 }
 
@@ -1002,7 +1020,7 @@ static int kvmppc_xive_native_create(struct kvm_device *dev, u32 type)
 	if (kvm->arch.xive)
 		return -EEXIST;
 
-	xive = kzalloc(sizeof(*xive), GFP_KERNEL);
+	xive = kvmppc_xive_get_device(kvm, type);
 	if (!xive)
 		return -ENOMEM;
 
@@ -1182,7 +1200,7 @@ struct kvm_device_ops kvm_xive_native_ops = {
 	.name = "kvm-xive-native",
 	.create = kvmppc_xive_native_create,
 	.init = kvmppc_xive_native_init,
-	.destroy = kvmppc_xive_native_free,
+	.release = kvmppc_xive_native_release,
 	.set_attr = kvmppc_xive_native_set_attr,
 	.get_attr = kvmppc_xive_native_get_attr,
 	.has_attr = kvmppc_xive_native_has_attr,
arch/powerpc/kvm/powerpc.c +9 -0
index f54926c78320..7f41a68d8094 100644
@@ -501,6 +501,15 @@ void kvm_arch_destroy_vm(struct kvm *kvm)
 
 	mutex_unlock(&kvm->lock);
 
+	/*
+	 * Free the XIVE devices which are not directly freed by the
+	 * device 'release' method
+	 */
+	kfree(kvm->arch.xive_devices.native);
+	kvm->arch.xive_devices.native = NULL;
+	kfree(kvm->arch.xive_devices.xics_on_xive);
+	kvm->arch.xive_devices.xics_on_xive = NULL;
+
 	/* drop the module reference */
 	module_put(kvm->arch.kvm_ops->owner);
 }
Signature · 1 line
2.20.1
Reviewed-by on v6 17/17
#20 lore raw

Re: [PATCH v6 17/17] KVM: PPC: Book3S HV: XIVE: replace the 'destroy' method by a 'release' method

Quoted 13 lines · Cédric Le Goater

When a P9 sPAPR VM boots, the CAS negotiation process determines which interrupt mode to use (XICS legacy or XIVE native) and invokes a machine reset to activate the chosen mode.

We introduce 'release' methods for the XICS-on-XIVE and the XIVE native KVM devices which are called when the file descriptor of the device is closed after the TIMA and ESB pages have been unmapped. They perform the necessary cleanups : clear the vCPU interrupt presenters that could be attached and then destroy the device. The 'release' methods replace the 'destroy' methods as 'destroy' is not called anymore once 'release' is. Compatibility with older QEMU is nevertheless maintained.

Quoted 8 lines

This is not considered as a safe operation as the vCPUs are still running and could be referencing the KVM device through their presenters. To protect the system from any breakage, the kvmppc_xive objects representing both KVM devices are now stored in an array under the VM. Allocation is performed on first usage and memory is freed only when the VM exits.

Quoted 268 lines

---

Changes since v5 :

- replaced the kvmppc_xive device array with named kvmppc_xive structs.
- in kvmppc_xive_cleanup_vcpu(), replaced xc->xive by vcpu->kvm->arch.xive which is always valid in this context
- in kvmppc_xive_release(), removed the comment on the race. The device being destroyed when the fd is closed, dev->private should always be valid when used. Also removed the kick_all_cpus_sync() to flush the IPIs.
- in the both release operation, removed the test on &kvm->online_vcpus.
- devices are now freed when the machine exits.

 arch/powerpc/include/asm/kvm_host.h   |  6 ++-
 arch/powerpc/kvm/book3s_xive.h        |  1 +
 arch/powerpc/kvm/book3s_xive.c        | 67 ++++++++++++++++++++++++---
 arch/powerpc/kvm/book3s_xive_native.c | 28 +++++++++--
 arch/powerpc/kvm/powerpc.c            |  9 ++++
 5 files changed, 99 insertions(+), 12 deletions(-)
arch/powerpc/include/asm/kvm_host.h +5 -1
index 9cc6abdce1b9..c7abc7c3c6cd 100644
@@ -313,7 +313,11 @@ struct kvm_arch {
 #endif
 #ifdef CONFIG_KVM_XICS
 	struct kvmppc_xics *xics;
-	struct kvmppc_xive *xive;
+	struct kvmppc_xive *xive;    /* Current XIVE device in use */
+	struct {
+		struct kvmppc_xive *native;
+		struct kvmppc_xive *xics_on_xive;
+	} xive_devices;
 	struct kvmppc_passthru_irqmap *pimap;
 #endif
 	struct kvmppc_ops *kvm_ops;
arch/powerpc/kvm/book3s_xive.h +1 -0
index e011622dc038..426146332984 100644
@@ -283,6 +283,7 @@ void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb);
 int kvmppc_xive_select_target(struct kvm *kvm, u32 *server, u8 prio);
 int kvmppc_xive_attach_escalation(struct kvm_vcpu *vcpu, u8 prio,
 				  bool single_escalation);
+struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type);
 
 #endif /* CONFIG_KVM_XICS */
 #endif /* _KVM_PPC_BOOK3S_XICS_H */
arch/powerpc/kvm/book3s_xive.c +61 -6
index 480a3fc6b9fd..922689b768e6 100644
@@ -1100,9 +1100,15 @@ void kvmppc_xive_disable_vcpu_interrupts(struct kvm_vcpu *vcpu)
 void kvmppc_xive_cleanup_vcpu(struct kvm_vcpu *vcpu)
 {
 	struct kvmppc_xive_vcpu *xc = vcpu->arch.xive_vcpu;
-	struct kvmppc_xive *xive = xc->xive;
+	struct kvmppc_xive *xive = vcpu->kvm->arch.xive;
 	int i;
 
+	if (!kvmppc_xics_enabled(vcpu))
+		return;
+
+	if (!xc)
+		return;
+
 	pr_devel("cleanup_vcpu(cpu=%d)\n", xc->server_num);
 
 	/* Ensure no interrupt is still routed to that VP */
@@ -1141,6 +1147,10 @@ void kvmppc_xive_cleanup_vcpu(struct kvm_vcpu *vcpu)
 	}
 	/* Free the VP */
 	kfree(xc);
+
+	/* Cleanup the vcpu */
+	vcpu->arch.irq_type = KVMPPC_IRQ_DEFAULT;
+	vcpu->arch.xive_vcpu = NULL;
 }
 
 int kvmppc_xive_connect_vcpu(struct kvm_device *dev,
@@ -1158,7 +1168,7 @@ int kvmppc_xive_connect_vcpu(struct kvm_device *dev,
 	}
 	if (xive->kvm != vcpu->kvm)
 		return -EPERM;
-	if (vcpu->arch.irq_type)
+	if (vcpu->arch.irq_type != KVMPPC_IRQ_DEFAULT)
 		return -EBUSY;
 	if (kvmppc_xive_find_server(vcpu->kvm, cpu)) {
 		pr_devel("Duplicate !\n");
@@ -1824,12 +1834,26 @@ void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb)
 	}
 }
 
-static void kvmppc_xive_free(struct kvm_device *dev)
+/*
+ * Called when device fd is closed
+ */
+static void kvmppc_xive_release(struct kvm_device *dev)
 {
 	struct kvmppc_xive *xive = dev->private;
 	struct kvm *kvm = xive->kvm;
+	struct kvm_vcpu *vcpu;
 	int i;
 
+	pr_devel("Releasing xive device\n");
+
+	/*
+	 * When releasing the KVM device fd, the vCPUs can still be
+	 * running and we should clean up the vCPU interrupt
+	 * presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm)
+		kvmppc_xive_cleanup_vcpu(vcpu);
+
 	debugfs_remove(xive->dentry);
 
 	if (kvm)
@@ -1846,11 +1870,42 @@ static void kvmppc_xive_free(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
+	/*
+	 * A reference of the kvmppc_xive pointer is now kept under
+	 * the xive_devices struct of the machine for reuse. It is
+	 * freed when the VM is destroyed for now until we fix all the
+	 * execution paths.
+	 */
 
-	kfree(xive);
 	kfree(dev);
 }
 
+/*
+ * When the guest chooses the interrupt mode (XICS legacy or XIVE
+ * native), the VM will switch of KVM device. The previous device will
+ * be "released" before the new one is created.
+ *
+ * Until we are sure all execution paths are well protected, provide a
+ * fail safe (transitional) method for device destruction, in which
+ * the XIVE device pointer is recycled and not directly freed.
+ */
+struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type)
+{
+	struct kvmppc_xive **kvm_xive_device = type == KVM_DEV_TYPE_XIVE ?
+		&kvm->arch.xive_devices.native :
+		&kvm->arch.xive_devices.xics_on_xive;
+	struct kvmppc_xive *xive = *kvm_xive_device;
+
+	if (!xive) {
+		xive = kzalloc(sizeof(*xive), GFP_KERNEL);
+		*kvm_xive_device = xive;
+	} else {
+		memset(xive, 0, sizeof(*xive));
+	}
+
+	return xive;
+}
+
 static int kvmppc_xive_create(struct kvm_device *dev, u32 type)
 {
 	struct kvmppc_xive *xive;
@@ -1859,7 +1914,7 @@ static int kvmppc_xive_create(struct kvm_device *dev, u32 type)
 
 	pr_devel("Creating xive for partition\n");
 
-	xive = kzalloc(sizeof(*xive), GFP_KERNEL);
+	xive = kvmppc_xive_get_device(kvm, type);
 	if (!xive)
 		return -ENOMEM;
 
@@ -2024,7 +2079,7 @@ struct kvm_device_ops kvm_xive_ops = {
 	.name = "kvm-xive",
 	.create = kvmppc_xive_create,
 	.init = kvmppc_xive_init,
-	.destroy = kvmppc_xive_free,
+	.release = kvmppc_xive_release,
 	.set_attr = xive_set_attr,
 	.get_attr = xive_get_attr,
 	.has_attr = xive_has_attr,
arch/powerpc/kvm/book3s_xive_native.c +23 -5
index 62648f833adf..0497272a72fa 100644
@@ -964,15 +964,27 @@ static int kvmppc_xive_native_has_attr(struct kvm_device *dev,
 	return -ENXIO;
 }
 
-static void kvmppc_xive_native_free(struct kvm_device *dev)
+/*
+ * Called when device fd is closed
+ */
+static void kvmppc_xive_native_release(struct kvm_device *dev)
 {
 	struct kvmppc_xive *xive = dev->private;
 	struct kvm *kvm = xive->kvm;
+	struct kvm_vcpu *vcpu;
 	int i;
 
 	debugfs_remove(xive->dentry);
 
-	pr_devel("Destroying xive native device\n");
+	pr_devel("Releasing xive native device\n");
+
+	/*
+	 * When releasing the KVM device fd, the vCPUs can still be
+	 * running and we should clean up the vCPU interrupt
+	 * presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm)
+		kvmppc_xive_native_cleanup_vcpu(vcpu);
 
 	if (kvm)
 		kvm->arch.xive = NULL;
@@ -987,7 +999,13 @@ static void kvmppc_xive_native_free(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
-	kfree(xive);
+	/*
+	 * A reference of the kvmppc_xive pointer is now kept under
+	 * the xive_devices struct of the machine for reuse. It is
+	 * freed when the VM is destroyed for now until we fix all the
+	 * execution paths.
+	 */
+
 	kfree(dev);
 }
 
@@ -1002,7 +1020,7 @@ static int kvmppc_xive_native_create(struct kvm_device *dev, u32 type)
 	if (kvm->arch.xive)
 		return -EEXIST;
 
-	xive = kzalloc(sizeof(*xive), GFP_KERNEL);
+	xive = kvmppc_xive_get_device(kvm, type);
 	if (!xive)
 		return -ENOMEM;
 
@@ -1182,7 +1200,7 @@ struct kvm_device_ops kvm_xive_native_ops = {
 	.name = "kvm-xive-native",
 	.create = kvmppc_xive_native_create,
 	.init = kvmppc_xive_native_init,
-	.destroy = kvmppc_xive_native_free,
+	.release = kvmppc_xive_native_release,
 	.set_attr = kvmppc_xive_native_set_attr,
 	.get_attr = kvmppc_xive_native_get_attr,
 	.has_attr = kvmppc_xive_native_has_attr,
arch/powerpc/kvm/powerpc.c +9 -0
index f54926c78320..7f41a68d8094 100644
@@ -501,6 +501,15 @@ void kvm_arch_destroy_vm(struct kvm *kvm)
 
 	mutex_unlock(&kvm->lock);
 
+	/*
+	 * Free the XIVE devices which are not directly freed by the
+	 * device 'release' method
+	 */
+	kfree(kvm->arch.xive_devices.native);
+	kvm->arch.xive_devices.native = NULL;
+	kfree(kvm->arch.xive_devices.xics_on_xive);
+	kvm->arch.xive_devices.xics_on_xive = NULL;
+
 	/* drop the module reference */
 	module_put(kvm->arch.kvm_ops->owner);
 }
Signature · 4 lines
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson
#21 lore raw

PATCH 1/1 KVM: PPC: Book3S HV: XIVE: Prevent races when releasing device

In reply to Cédric Le Goater

Now that we have the possibility of a XIVE or XICS-on-XIVE device being released while the VM is still running, we need to be careful about races and potential use-after-free bugs. Although the kvmppc_xive struct is not freed, but kept around for re-use, the kvmppc_xive_vcpu structs are freed, and they are used extensively in both the XIVE native and XICS-on-XIVE code.

There are various ways in which XIVE code gets invoked:

- VCPU entry and exit, which do push and pull operations on the XIVE hardware
- one_reg get and set functions (vcpu->mutex is held)
- XICS hypercalls (but only inside guest execution, not from kvmppc_pseries_do_hcall)
- device creation calls (kvm->lock is held)
- device callbacks - get/set attribute, mmap, pagefault, release/destroy
- set_mapped/clr_mapped calls (kvm->lock is held)
- connect_vcpu calls
- debugfs file read callbacks

Inside a device release function, we know that userspace cannot have an open file descriptor referring to the device, nor can it have any mmapped regions from the device. Therefore the device callbacks are excluded, as are the connect_vcpu calls (since they need a fd for the device). Further, since the caller holds the kvm->lock mutex, no other device creation calls or set/clr_mapped calls can be executing concurrently.

To exclude VCPU execution and XICS hypercalls, we temporarily set kvm->arch.mmu_ready to 0. This forces any VCPU task that is trying to enter the guest to take the kvm->lock mutex, which is held by the caller of the release function. Then, sending an IPI to all other CPUs forces any VCPU currently executing in the guest to exit.

Finally, we take the vcpu->mutex for each VCPU around the process of cleaning up and freeing its XIVE data structures, in order to exclude any one_reg get/set calls.

This patch does not address the races in the debugfs read callbacks. They will be addressed in a later patch.

---
Cedric,

I think we need this on top of your patch 17/17, since you are freeing the kvmppc_xive_vcpu structs. With this, we can probably also free the kvmppc_xive struct, though I haven't yet completely satisfied myself about that. If you can see any entry point that I have missed (other than the debugfs stuff), let me know.

I can either apply this on top of your series, or fold it into your 17/17, whichever you prefer.

 arch/powerpc/kvm/book3s_xive.c        | 44 ++++++++++++++++++++++++++++++-----
 arch/powerpc/kvm/book3s_xive_native.c | 40 +++++++++++++++++++++++++++----
 2 files changed, 74 insertions(+), 10 deletions(-)
arch/powerpc/kvm/book3s_xive.c +38 -6
index 922689b..ce5196e 100644
@@ -846,7 +846,8 @@ int kvmppc_xive_set_icp(struct kvm_vcpu *vcpu, u64 icpval)
 
 	/*
 	 * We can't update the state of a "pushed" VCPU, but that
-	 * shouldn't happen.
+	 * shouldn't happen because the vcpu->mutex makes running a
+	 * vcpu mutually exclusive with doing one_reg get/set on it.
 	 */
 	if (WARN_ON(vcpu->arch.xive_pushed))
 		return -EIO;
@@ -1835,7 +1836,7 @@ void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb)
 }
 
 /*
- * Called when device fd is closed
+ * Called when device fd is closed.  kvm->lock is held.
  */
 static void kvmppc_xive_release(struct kvm_device *dev)
 {
@@ -1843,16 +1844,42 @@ static void kvmppc_xive_release(struct kvm_device *dev)
 	struct kvm *kvm = xive->kvm;
 	struct kvm_vcpu *vcpu;
 	int i;
+	int was_ready;
 
 	pr_devel("Releasing xive device\n");
 
 	/*
-	 * When releasing the KVM device fd, the vCPUs can still be
-	 * running and we should clean up the vCPU interrupt
-	 * presenters first.
+	 * Clearing mmu_ready temporarily while holding kvm->lock
+	 * is a way of ensuring that no vcpus can enter the guest
+	 * until we drop kvm->lock.  Doing kick_all_cpus_sync()
+	 * ensures that any vcpu executing inside the guest has
+	 * exited the guest.  Once kick_all_cpus_sync() has finished,
+	 * we know that no vcpu can be executing the XIVE push or
+	 * pull code, or executing a XICS hcall.
+	 *
+	 * Since this is the device release function, we know that
+	 * userspace does not have any open fd referring to the
+	 * device.  Therefore there can not be any of the device
+	 * attribute set/get functions being executed concurrently,
+	 * and similarly, the connect_vcpu and set/clr_mapped
+	 * functions also cannot be being executed.
+	 */
+	was_ready = kvm->arch.mmu_ready;
+	kvm->arch.mmu_ready = 0;
+	kick_all_cpus_sync();
+
+	/*
+	 * We should clean up the vCPU interrupt presenters first.
 	 */
-	kvm_for_each_vcpu(i, vcpu, kvm)
+	kvm_for_each_vcpu(i, vcpu, kvm) {
+		/*
+		 * Take vcpu->mutex to ensure that no one_reg get/set ioctl
+		 * (i.e. kvmppc_xive_[gs]et_icp) can be done concurrently.
+		 */
+		mutex_lock(&vcpu->mutex);
 		kvmppc_xive_cleanup_vcpu(vcpu);
+		mutex_unlock(&vcpu->mutex);
+	}
 
 	debugfs_remove(xive->dentry);
 
@@ -1870,6 +1897,8 @@ static void kvmppc_xive_release(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
+	kvm->arch.mmu_ready = was_ready;
+
 	/*
 	 * A reference of the kvmppc_xive pointer is now kept under
 	 * the xive_devices struct of the machine for reuse. It is
@@ -1906,6 +1935,9 @@ struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type)
 	return xive;
 }
 
+/*
+ * Create a XICS device with XIVE backend.  kvm->lock is held.
+ */
 static int kvmppc_xive_create(struct kvm_device *dev, u32 type)
 {
 	struct kvmppc_xive *xive;
arch/powerpc/kvm/book3s_xive_native.c +36 -4
index 0497272a..3416c82 100644
@@ -973,18 +973,45 @@ static void kvmppc_xive_native_release(struct kvm_device *dev)
 	struct kvm *kvm = xive->kvm;
 	struct kvm_vcpu *vcpu;
 	int i;
+	int was_ready;
 
 	debugfs_remove(xive->dentry);
 
 	pr_devel("Releasing xive native device\n");
 
 	/*
-	 * When releasing the KVM device fd, the vCPUs can still be
-	 * running and we should clean up the vCPU interrupt
-	 * presenters first.
+	 * Clearing mmu_ready temporarily while holding kvm->lock
+	 * is a way of ensuring that no vcpus can enter the guest
+	 * until we drop kvm->lock.  Doing kick_all_cpus_sync()
+	 * ensures that any vcpu executing inside the guest has
+	 * exited the guest.  Once kick_all_cpus_sync() has finished,
+	 * we know that no vcpu can be executing the XIVE push or
+	 * pull code or accessing the XIVE MMIO regions.
+	 *
+	 * Since this is the device release function, we know that
+	 * userspace does not have any open fd or mmap referring to
+	 * the device.  Therefore there can not be any of the
+	 * device attribute set/get, mmap, or page fault functions
+	 * being executed concurrently, and similarly, the
+	 * connect_vcpu and set/clr_mapped functions also cannot
+	 * be being executed.
 	 */
-	kvm_for_each_vcpu(i, vcpu, kvm)
+	was_ready = kvm->arch.mmu_ready;
+	kvm->arch.mmu_ready = 0;
+	kick_all_cpus_sync();
+
+	/*
+	 * We should clean up the vCPU interrupt presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm) {
+		/*
+		 * Take vcpu->mutex to ensure that no one_reg get/set ioctl
+		 * (i.e. kvmppc_xive_native_[gs]et_vp) can be being done.
+		 */
+		mutex_lock(&vcpu->mutex);
 		kvmppc_xive_native_cleanup_vcpu(vcpu);
+		mutex_unlock(&vcpu->mutex);
+	}
 
 	if (kvm)
 		kvm->arch.xive = NULL;
@@ -999,6 +1026,8 @@ static void kvmppc_xive_native_release(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
+	kvm->arch.mmu_ready = was_ready;
+
 	/*
 	 * A reference of the kvmppc_xive pointer is now kept under
 	 * the xive_devices struct of the machine for reuse. It is
@@ -1009,6 +1038,9 @@ static void kvmppc_xive_native_release(struct kvm_device *dev)
 	kfree(dev);
 }
 
+/*
+ * Create a XIVE device.  kvm->lock is held.
+ */
 static int kvmppc_xive_native_create(struct kvm_device *dev, u32 type)
 {
 	struct kvmppc_xive *xive;
Signature · 1 line
2.7.4
#22 lore raw

PATCH v2 1/1 KVM: PPC: Book3S HV: XIVE: Prevent races when releasing device

Now that we have the possibility of a XIVE or XICS-on-XIVE device being released while the VM is still running, we need to be careful about races and potential use-after-free bugs. Although the kvmppc_xive struct is not freed, but kept around for re-use, the kvmppc_xive_vcpu structs are freed, and they are used extensively in both the XIVE native and XICS-on-XIVE code.

There are various ways in which XIVE code gets invoked:

- VCPU entry and exit, which do push and pull operations on the XIVE hardware
- one_reg get and set functions (vcpu->mutex is held)
- XICS hypercalls (but only inside guest execution, not from kvmppc_pseries_do_hcall)
- device creation calls (kvm->lock is held)
- device callbacks - get/set attribute, mmap, pagefault, release/destroy
- set_mapped/clr_mapped calls (kvm->lock is held)
- connect_vcpu calls
- debugfs file read callbacks

Inside a device release function, we know that userspace cannot have an open file descriptor referring to the device, nor can it have any mmapped regions from the device. Therefore the device callbacks are excluded, as are the connect_vcpu calls (since they need a fd for the device). Further, since the caller holds the kvm->lock mutex, no other device creation calls or set/clr_mapped calls can be executing concurrently.

To exclude VCPU execution and XICS hypercalls, we temporarily set kvm->arch.mmu_ready to 0. This forces any VCPU task that is trying to enter the guest to take the kvm->lock mutex, which is held by the caller of the release function. Then, sending an IPI to all other CPUs forces any VCPU currently executing in the guest to exit.

Finally, we take the vcpu->mutex for each VCPU around the process of cleaning up and freeing its XIVE data structures, in order to exclude any one_reg get/set calls.

To exclude the debugfs read callbacks, we just need to ensure that debugfs_remove is called before freeing any data structures. Once it returns we know that no CPU can be executing the callbacks (for our kvmppc_xive instance).

---
v2: move debugfs_remove call to eliminate race

 arch/powerpc/kvm/book3s_xive.c        | 51 ++++++++++++++++++++++++++++-------
 arch/powerpc/kvm/book3s_xive_native.c | 43 ++++++++++++++++++++++++-----
 2 files changed, 78 insertions(+), 16 deletions(-)
arch/powerpc/kvm/book3s_xive.c +41 -10
index 922689b..4280cd8 100644
@@ -846,7 +846,8 @@ int kvmppc_xive_set_icp(struct kvm_vcpu *vcpu, u64 icpval)
 
 	/*
 	 * We can't update the state of a "pushed" VCPU, but that
-	 * shouldn't happen.
+	 * shouldn't happen because the vcpu->mutex makes running a
+	 * vcpu mutually exclusive with doing one_reg get/set on it.
 	 */
 	if (WARN_ON(vcpu->arch.xive_pushed))
 		return -EIO;
@@ -1835,7 +1836,7 @@ void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb)
 }
 
 /*
- * Called when device fd is closed
+ * Called when device fd is closed.  kvm->lock is held.
  */
 static void kvmppc_xive_release(struct kvm_device *dev)
 {
@@ -1843,21 +1844,46 @@ static void kvmppc_xive_release(struct kvm_device *dev)
 	struct kvm *kvm = xive->kvm;
 	struct kvm_vcpu *vcpu;
 	int i;
+	int was_ready;
 
 	pr_devel("Releasing xive device\n");
 
+	debugfs_remove(xive->dentry);
+
 	/*
-	 * When releasing the KVM device fd, the vCPUs can still be
-	 * running and we should clean up the vCPU interrupt
-	 * presenters first.
+	 * Clearing mmu_ready temporarily while holding kvm->lock
+	 * is a way of ensuring that no vcpus can enter the guest
+	 * until we drop kvm->lock.  Doing kick_all_cpus_sync()
+	 * ensures that any vcpu executing inside the guest has
+	 * exited the guest.  Once kick_all_cpus_sync() has finished,
+	 * we know that no vcpu can be executing the XIVE push or
+	 * pull code, or executing a XICS hcall.
+	 *
+	 * Since this is the device release function, we know that
+	 * userspace does not have any open fd referring to the
+	 * device.  Therefore there can not be any of the device
+	 * attribute set/get functions being executed concurrently,
+	 * and similarly, the connect_vcpu and set/clr_mapped
+	 * functions also cannot be being executed.
 	 */
-	kvm_for_each_vcpu(i, vcpu, kvm)
-		kvmppc_xive_cleanup_vcpu(vcpu);
+	was_ready = kvm->arch.mmu_ready;
+	kvm->arch.mmu_ready = 0;
+	kick_all_cpus_sync();
 
-	debugfs_remove(xive->dentry);
+	/*
+	 * We should clean up the vCPU interrupt presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm) {
+		/*
+		 * Take vcpu->mutex to ensure that no one_reg get/set ioctl
+		 * (i.e. kvmppc_xive_[gs]et_icp) can be done concurrently.
+		 */
+		mutex_lock(&vcpu->mutex);
+		kvmppc_xive_cleanup_vcpu(vcpu);
+		mutex_unlock(&vcpu->mutex);
+	}
 
-	if (kvm)
-		kvm->arch.xive = NULL;
+	kvm->arch.xive = NULL;
 
 	/* Mask and free interrupts */
 	for (i = 0; i <= xive->max_sbid; i++) {
@@ -1870,6 +1896,8 @@ static void kvmppc_xive_release(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
+	kvm->arch.mmu_ready = was_ready;
+
 	/*
 	 * A reference of the kvmppc_xive pointer is now kept under
 	 * the xive_devices struct of the machine for reuse. It is
@@ -1906,6 +1934,9 @@ struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type)
 	return xive;
 }
 
+/*
+ * Create a XICS device with XIVE backend.  kvm->lock is held.
+ */
 static int kvmppc_xive_create(struct kvm_device *dev, u32 type)
 {
 	struct kvmppc_xive *xive;
arch/powerpc/kvm/book3s_xive_native.c +37 -6

70 lines not on this page · raw message

Signature · 1 line
2.7.4
author Reviewed-by on v2 1/1
#23 lore raw

Re: [PATCH v2] KVM: PPC: Book3S HV: XIVE: Prevent races when releasing device

Quoted 26 lines · Paul Mackerras

Now that we have the possibility of a XIVE or XICS-on-XIVE device being released while the VM is still running, we need to be careful about races and potential use-after-free bugs. Although the kvmppc_xive struct is not freed, but kept around for re-use, the kvmppc_xive_vcpu structs are freed, and they are used extensively in both the XIVE native and XICS-on-XIVE code.

There are various ways in which XIVE code gets invoked:

- VCPU entry and exit, which do push and pull operations on the XIVE hardware
- one_reg get and set functions (vcpu->mutex is held)
- XICS hypercalls (but only inside guest execution, not from kvmppc_pseries_do_hcall)
- device creation calls (kvm->lock is held)
- device callbacks - get/set attribute, mmap, pagefault, release/destroy
- set_mapped/clr_mapped calls (kvm->lock is held)
- connect_vcpu calls
- debugfs file read callbacks

Inside a device release function, we know that userspace cannot have an open file descriptor referring to the device, nor can it have any mmapped regions from the device. Therefore the device callbacks are excluded, as are the connect_vcpu calls (since they need a fd for the device). Further, since the caller holds the kvm->lock mutex, no other device creation calls or set/clr_mapped calls can be executing concurrently.

Quoted 5 lines

To exclude VCPU execution and XICS hypercalls, we temporarily set kvm->arch.mmu_ready to 0. This forces any VCPU task that is trying to enter the guest to take the kvm->lock mutex, which is held by the caller of the release function. Then, sending an IPI to all other CPUs forces any VCPU currently executing in the guest to exit.

For my understanding, this method is faster than looping on all vCPUs and calling kvm_vcpu_kick() ?

Quoted 11 lines

Finally, we take the vcpu->mutex for each VCPU around the process of cleaning up and freeing its XIVE data structures, in order to exclude any one_reg get/set calls.

To exclude the debugfs read callbacks, we just need to ensure that debugfs_remove is called before freeing any data structures. Once it returns we know that no CPU can be executing the callbacks (for our kvmppc_xive instance).

LGTM,

one minor comment below,

Thanks,

C.

arch/powerpc/kvm/book3s_xive_native.c

Quoted 108 lines

---
v2: move debugfs_remove call to eliminate race

 arch/powerpc/kvm/book3s_xive.c        | 51 ++++++++++++++++++++++++++++-------
 arch/powerpc/kvm/book3s_xive_native.c | 43 ++++++++++++++++++++++++-----
 2 files changed, 78 insertions(+), 16 deletions(-)
arch/powerpc/kvm/book3s_xive.c +41 -10
index 922689b..4280cd8 100644
@@ -846,7 +846,8 @@ int kvmppc_xive_set_icp(struct kvm_vcpu *vcpu, u64 icpval)
 
 	/*
 	 * We can't update the state of a "pushed" VCPU, but that
-	 * shouldn't happen.
+	 * shouldn't happen because the vcpu->mutex makes running a
+	 * vcpu mutually exclusive with doing one_reg get/set on it.
 	 */
 	if (WARN_ON(vcpu->arch.xive_pushed))
 		return -EIO;
@@ -1835,7 +1836,7 @@ void kvmppc_xive_free_sources(struct kvmppc_xive_src_block *sb)
 }
 
 /*
- * Called when device fd is closed
+ * Called when device fd is closed.  kvm->lock is held.
  */
 static void kvmppc_xive_release(struct kvm_device *dev)
 {
@@ -1843,21 +1844,46 @@ static void kvmppc_xive_release(struct kvm_device *dev)
 	struct kvm *kvm = xive->kvm;
 	struct kvm_vcpu *vcpu;
 	int i;
+	int was_ready;
 
 	pr_devel("Releasing xive device\n");
 
+	debugfs_remove(xive->dentry);
+
 	/*
-	 * When releasing the KVM device fd, the vCPUs can still be
-	 * running and we should clean up the vCPU interrupt
-	 * presenters first.
+	 * Clearing mmu_ready temporarily while holding kvm->lock
+	 * is a way of ensuring that no vcpus can enter the guest
+	 * until we drop kvm->lock.  Doing kick_all_cpus_sync()
+	 * ensures that any vcpu executing inside the guest has
+	 * exited the guest.  Once kick_all_cpus_sync() has finished,
+	 * we know that no vcpu can be executing the XIVE push or
+	 * pull code, or executing a XICS hcall.
+	 *
+	 * Since this is the device release function, we know that
+	 * userspace does not have any open fd referring to the
+	 * device.  Therefore there can not be any of the device
+	 * attribute set/get functions being executed concurrently,
+	 * and similarly, the connect_vcpu and set/clr_mapped
+	 * functions also cannot be being executed.
 	 */
-	kvm_for_each_vcpu(i, vcpu, kvm)
-		kvmppc_xive_cleanup_vcpu(vcpu);
+	was_ready = kvm->arch.mmu_ready;
+	kvm->arch.mmu_ready = 0;
+	kick_all_cpus_sync();
 
-	debugfs_remove(xive->dentry);
+	/*
+	 * We should clean up the vCPU interrupt presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm) {
+		/*
+		 * Take vcpu->mutex to ensure that no one_reg get/set ioctl
+		 * (i.e. kvmppc_xive_[gs]et_icp) can be done concurrently.
+		 */
+		mutex_lock(&vcpu->mutex);
+		kvmppc_xive_cleanup_vcpu(vcpu);
+		mutex_unlock(&vcpu->mutex);
+	}
 
-	if (kvm)
-		kvm->arch.xive = NULL;
+	kvm->arch.xive = NULL;
 
 	/* Mask and free interrupts */
 	for (i = 0; i <= xive->max_sbid; i++) {
@@ -1870,6 +1896,8 @@ static void kvmppc_xive_release(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
+	kvm->arch.mmu_ready = was_ready;
+
 	/*
 	 * A reference of the kvmppc_xive pointer is now kept under
 	 * the xive_devices struct of the machine for reuse. It is
@@ -1906,6 +1934,9 @@ struct kvmppc_xive *kvmppc_xive_get_device(struct kvm *kvm, u32 type)
 	return xive;
 }
 
+/*
+ * Create a XICS device with XIVE backend.  kvm->lock is held.
+ */
 static int kvmppc_xive_create(struct kvm_device *dev, u32 type)
 {
 	struct kvmppc_xive *xive;
arch/powerpc/kvm/book3s_xive_native.c +0 -0
index 0497272a..5e14df1 100644

May be also add the comment :

kvm->lock is held when kvmppc_xive_native_release() is called.

Quoted 74 lines
arch/powerpc/kvm/book3s_xive_native.c +37 -6
@@ -973,21 +973,47 @@ static void kvmppc_xive_native_release(struct kvm_device *dev)
 	struct kvm *kvm = xive->kvm;
 	struct kvm_vcpu *vcpu;
 	int i;
+	int was_ready;
 
 	debugfs_remove(xive->dentry);
 
 	pr_devel("Releasing xive native device\n");
 
 	/*
-	 * When releasing the KVM device fd, the vCPUs can still be
-	 * running and we should clean up the vCPU interrupt
-	 * presenters first.
+	 * Clearing mmu_ready temporarily while holding kvm->lock
+	 * is a way of ensuring that no vcpus can enter the guest
+	 * until we drop kvm->lock.  Doing kick_all_cpus_sync()
+	 * ensures that any vcpu executing inside the guest has
+	 * exited the guest.  Once kick_all_cpus_sync() has finished,
+	 * we know that no vcpu can be executing the XIVE push or
+	 * pull code or accessing the XIVE MMIO regions.
+	 *
+	 * Since this is the device release function, we know that
+	 * userspace does not have any open fd or mmap referring to
+	 * the device.  Therefore there can not be any of the
+	 * device attribute set/get, mmap, or page fault functions
+	 * being executed concurrently, and similarly, the
+	 * connect_vcpu and set/clr_mapped functions also cannot
+	 * be being executed.
 	 */
-	kvm_for_each_vcpu(i, vcpu, kvm)
+	was_ready = kvm->arch.mmu_ready;
+	kvm->arch.mmu_ready = 0;
+	kick_all_cpus_sync();
+
+	/*
+	 * We should clean up the vCPU interrupt presenters first.
+	 */
+	kvm_for_each_vcpu(i, vcpu, kvm) {
+		/*
+		 * Take vcpu->mutex to ensure that no one_reg get/set ioctl
+		 * (i.e. kvmppc_xive_native_[gs]et_vp) can be being done.
+		 */
+		mutex_lock(&vcpu->mutex);
 		kvmppc_xive_native_cleanup_vcpu(vcpu);
+		mutex_unlock(&vcpu->mutex);
+	}
 
-	if (kvm)
-		kvm->arch.xive = NULL;
+	kvm->arch.xive = NULL;
 
 	for (i = 0; i <= xive->max_sbid; i++) {
 		if (xive->src_blocks[i])
@@ -999,6 +1025,8 @@ static void kvmppc_xive_native_release(struct kvm_device *dev)
 	if (xive->vp_base != XIVE_INVALID_VP)
 		xive_native_free_vp_block(xive->vp_base);
 
+	kvm->arch.mmu_ready = was_ready;
+
 	/*
 	 * A reference of the kvmppc_xive pointer is now kept under
 	 * the xive_devices struct of the machine for reuse. It is
@@ -1009,6 +1037,9 @@ static void kvmppc_xive_native_release(struct kvm_device *dev)
 	kfree(dev);
 }
 
+/*
+ * Create a XIVE device.  kvm->lock is held.
+ */
 static int kvmppc_xive_native_create(struct kvm_device *dev, u32 type)
 {
 	struct kvmppc_xive *xive;
#24 lore raw
In reply to Cédric Le Goater
Quoted 40 lines · Cédric Le Goater

On the POWER9 processor, the XIVE interrupt controller can control interrupt sources using MMIOs to trigger events, to EOI or to turn off the sources. Priority management and interrupt acknowledgment is also controlled by MMIO in the CPU presenter sub-engine.

PowerNV/baremetal Linux runs natively under XIVE but sPAPR guests need special support from the hypervisor to do the same. This is called the XIVE native exploitation mode and today, it can be activated under the PowerPC Hypervisor, pHyp. However, Linux/KVM lacks XIVE native support and still offers the old interrupt mode interface using a KVM device implementing the XICS hcalls over XIVE.

The following series is proposal to add the same support under KVM.

A new KVM device is introduced for the XIVE native exploitation mode. It reuses most of the XICS-over-XIVE glue implementation structures which are internal to KVM but has a completely different interface. A set of KVM device ioctls provide support for the hypervisor calls, all handled in QEMU, to configure the sources and the event queues. From there, all interrupt control is transferred to the guest which can use MMIOs.

These MMIO regions (ESB and TIMA) are exposed to guests in QEMU, similarly to VFIO, and the associated VMAs are populated dynamically with the appropriate pages using a fault handler. These are now implemented using mmap()s of the KVM device fd.

Migration has its own specific needs regarding memory. The patchset provides a specific control to quiesce XIVE before capturing the memory. The save and restore of the internal state is based on the same ioctls used for the hcalls.

On a POWER9 sPAPR machine, the Client Architecture Support (CAS) negotiation process determines whether the guest operates with a interrupt controller using the XICS legacy model, as found on POWER8, or in XIVE exploitation mode. Which means that the KVM interrupt device should be created at run-time, after the machine has started. This requires extra support from KVM to destroy KVM devices. It is introduced at the end of the patchset and requires some attention.

Quoted 2 lines

This is based on Linux 5.1-rc5 and is a candidate for 5.2. The OPAL patches have been merged now.

Thanks, patch series applied to my kvm-ppc-next tree. I added two patches of mine on top to make sure we exclude other execution paths in the device release method, and to clear the escalation interrupt hardware pointers on release. I also modified your last patch to free the xive structures in book3s.c rather than powerpc.c in order to fix compilation for Book E configs.

Paul.

#25 lore raw
Quoted 44 lines · Paul Mackerras
Quoted 40 lines · Cédric Le Goater

On the POWER9 processor, the XIVE interrupt controller can control interrupt sources using MMIOs to trigger events, to EOI or to turn off the sources. Priority management and interrupt acknowledgment is also controlled by MMIO in the CPU presenter sub-engine.

PowerNV/baremetal Linux runs natively under XIVE but sPAPR guests need special support from the hypervisor to do the same. This is called the XIVE native exploitation mode and today, it can be activated under the PowerPC Hypervisor, pHyp. However, Linux/KVM lacks XIVE native support and still offers the old interrupt mode interface using a KVM device implementing the XICS hcalls over XIVE.

The following series is proposal to add the same support under KVM.

A new KVM device is introduced for the XIVE native exploitation mode. It reuses most of the XICS-over-XIVE glue implementation structures which are internal to KVM but has a completely different interface. A set of KVM device ioctls provide support for the hypervisor calls, all handled in QEMU, to configure the sources and the event queues. From there, all interrupt control is transferred to the guest which can use MMIOs.

These MMIO regions (ESB and TIMA) are exposed to guests in QEMU, similarly to VFIO, and the associated VMAs are populated dynamically with the appropriate pages using a fault handler. These are now implemented using mmap()s of the KVM device fd.

Migration has its own specific needs regarding memory. The patchset provides a specific control to quiesce XIVE before capturing the memory. The save and restore of the internal state is based on the same ioctls used for the hcalls.

On a POWER9 sPAPR machine, the Client Architecture Support (CAS) negotiation process determines whether the guest operates with a interrupt controller using the XICS legacy model, as found on POWER8, or in XIVE exploitation mode. Which means that the KVM interrupt device should be created at run-time, after the machine has started. This requires extra support from KVM to destroy KVM devices. It is introduced at the end of the patchset and requires some attention.

Quoted 2 lines

This is based on Linux 5.1-rc5 and is a candidate for 5.2. The OPAL patches have been merged now.

Quoted 6 lines

Thanks, patch series applied to my kvm-ppc-next tree. I added two patches of mine on top to make sure we exclude other execution paths in the device release method, and to clear the escalation interrupt hardware pointers on release. I also modified your last patch to free the xive structures in book3s.c rather than powerpc.c in order to fix compilation for Book E configs.

OK. I have one minor cleanup removing bogus checks in the release method of the KVM device.

Thanks,

C.