Skip to contentSkip to conversation

spapr: add KVM support to the XIVE interrupt mode

PATCH Superseded Cédric Le Goater 2 participants 32 messages Feb 22, 2019 – Mar 11, 2019

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover spapr: add KVM support to the XIVE interrupt mode

Hello,

This is the v2 of the QEMU/KVM patchset taking into account the remarks on the interface with Linux/KVM.

The first patches introduce the XIVE KVM device, state synchronization and migration support under KVM. The second part of the patchset modifies the XICS and XIVE interrupt models to add KVM support to the 'dual' IRQ backend.

GitHub trees available here :

QEMU sPAPR:

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

Linux/KVM:

https://github.com/legoater/linux/commits/xive-5.0

OPAL:

https://github.com/legoater/skiboot/commits/xive

Thanks,

C.

Change since v1:

- Reworked most of the KVM interface
- Reworked *All* hcalls which are now handled at the QEMU level, possibly extended with a KVM device ioctl when required.
- TIMA and ESB special mapping done on the KVM device fd.
- Tested on nested
- Implemented the device fallback mode when a kernel_irqchip is not available and not required. Useful on nested to use XIVE.
- Fix device hotplug when VM is stopped (Is this necessary ?)

Cédric Le Goater (13):
spapr/xive: add KVM support
spapr/xive: add hcall support when under KVM spapr/xive: activate KVM support
spapr/xive: add state synchronization with KVM spapr/xive: introduce a VM state change handler spapr/xive: add migration support for KVM
spapr/xive: fix migration of the XiveTCTX under TCG spapr/rtas: modify spapr_rtas_register() to remove RTAS handlers
sysbus: add a sysbus_mmio_unmap() helper
spapr: introduce routines to delete the KVM IRQ device
spapr: check for the activation of the KVM IRQ device
spapr: add KVM support to the 'dual' machine spapr/xive: fix device hotplug when VM is stopped

 default-configs/ppc64-softmmu.mak |   1 +
 include/hw/ppc/spapr.h            |   4 +
 include/hw/ppc/spapr_xive.h       |  39 ++
 include/hw/ppc/xics_spapr.h       |   1 +
 include/hw/ppc/xive.h             |  15 +
 include/hw/sysbus.h               |   1 +
 target/ppc/kvm_ppc.h              |   6 +
 hw/core/sysbus.c                  |  10 +
 hw/intc/spapr_xive.c              | 193 ++++++-
 hw/intc/spapr_xive_kvm.c          | 807 ++++++++++++++++++++++++++++++
 hw/intc/xics_kvm.c                | 108 +++-
 hw/intc/xive.c                    |  44 +-
 hw/ppc/spapr_irq.c                | 136 +++--
 hw/ppc/spapr_rtas.c               |   2 +-
 target/ppc/kvm.c                  |   7 +
 hw/intc/Makefile.objs             |   1 +
 16 files changed, 1315 insertions(+), 60 deletions(-)
 create mode 100644 hw/intc/spapr_xive_kvm.c
Signature · 1 line
2.20.1
#1 lore raw

PATCH v2 1/13 spapr/xive: add KVM support

This introduces a set of helpers when KVM is in use, which create the KVM XIVE device, initialize the interrupt sources at a KVM level and connect the interrupt presenters to the vCPU.

They also handle the initialization of the TIMA and the source ESB memory regions of the controller. These have a different type under KVM. They are 'ram device' memory mappings, similarly to VFIO, exposed to the guest and the associated VMAs on the host are populated dynamically with the appropriate pages using a fault handler.

---
 default-configs/ppc64-softmmu.mak |   1 +
 include/hw/ppc/spapr_xive.h       |  10 ++
 include/hw/ppc/xive.h             |  13 ++
 target/ppc/kvm_ppc.h              |   6 +
 hw/intc/spapr_xive.c              |  48 +++++-
 hw/intc/spapr_xive_kvm.c          | 237 ++++++++++++++++++++++++++++++
 hw/intc/xive.c                    |  21 ++-
 hw/ppc/spapr_irq.c                |   6 +-
 target/ppc/kvm.c                  |   7 +
 hw/intc/Makefile.objs             |   1 +
 10 files changed, 340 insertions(+), 10 deletions(-)
 create mode 100644 hw/intc/spapr_xive_kvm.c
default-configs/ppc64-softmmu.mak +1 -0
index 7f34ad0528ed..c1bf5cd951f5 100644
@@ -18,6 +18,7 @@ CONFIG_XICS_SPAPR=$(CONFIG_PSERIES)
 CONFIG_XICS_KVM=$(call land,$(CONFIG_PSERIES),$(CONFIG_KVM))
 CONFIG_XIVE=$(CONFIG_PSERIES)
 CONFIG_XIVE_SPAPR=$(CONFIG_PSERIES)
+CONFIG_XIVE_KVM=$(call land,$(CONFIG_PSERIES),$(CONFIG_KVM))
 CONFIG_MEM_DEVICE=y
 CONFIG_DIMM=y
 CONFIG_SPAPR_RNG=y
include/hw/ppc/spapr_xive.h +10 -0
index 2d31f24e3bfe..ab6732b14a02 100644
@@ -38,6 +38,10 @@ typedef struct sPAPRXive {
     /* TIMA mapping address */
     hwaddr        tm_base;
     MemoryRegion  tm_mmio;
+
+    /* KVM support */
+    int           fd;
+    void          *tm_mmap;
 } sPAPRXive;
 
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
@@ -49,5 +53,11 @@ void spapr_dt_xive(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
                    uint32_t phandle);
 void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx);
 void spapr_xive_mmio_set_enabled(sPAPRXive *xive, bool enable);
+void spapr_xive_map_mmio(sPAPRXive *xive);
+
+/*
+ * KVM XIVE device helpers
+ */
+void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +13 -0
index 13a487527b11..061d43fea24d 100644
@@ -140,6 +140,7 @@
 #ifndef PPC_XIVE_H
 #define PPC_XIVE_H
 
+#include "sysemu/kvm.h"
 #include "hw/qdev-core.h"
 #include "hw/sysbus.h"
 #include "hw/ppc/xive_regs.h"
@@ -194,6 +195,9 @@ typedef struct XiveSource {
     uint32_t        esb_shift;
     MemoryRegion    esb_mmio;
 
+    /* KVM support */
+    void            *esb_mmap;
+
     XiveNotifier    *xive;
 } XiveSource;
 
@@ -419,4 +423,13 @@ static inline uint32_t xive_nvt_cam_line(uint8_t nvt_blk, uint32_t nvt_idx)
     return (nvt_blk << 19) | nvt_idx;
 }
 
+/*
+ * KVM XIVE device helpers
+ */
+
+void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp);
+void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
+void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
+void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
+
 #endif /* PPC_XIVE_H */
target/ppc/kvm_ppc.h +6 -0
index bdfaa4e70a83..d2159660f9f2 100644
@@ -59,6 +59,7 @@ bool kvmppc_has_cap_fixup_hcalls(void);
 bool kvmppc_has_cap_htm(void);
 bool kvmppc_has_cap_mmu_radix(void);
 bool kvmppc_has_cap_mmu_hash_v3(void);
+bool kvmppc_has_cap_xive(void);
 int kvmppc_get_cap_safe_cache(void);
 int kvmppc_get_cap_safe_bounds_check(void);
 int kvmppc_get_cap_safe_indirect_branch(void);
@@ -307,6 +308,11 @@ static inline bool kvmppc_has_cap_mmu_hash_v3(void)
     return false;
 }
 
+static inline bool kvmppc_has_cap_xive(void)
+{
+    return false;
+}
+
 static inline int kvmppc_get_cap_safe_cache(void)
 {
     return 0;
hw/intc/spapr_xive.c +42 -6

84 lines not on this page · raw message

hw/intc/spapr_xive_kvm.c new file +237 -0

237 lines not on this page · raw message

hw/intc/xive.c +18 -3
index daa7badc8492..0284b5803551 100644
@@ -491,6 +491,15 @@ static void xive_tctx_realize(DeviceState *dev, Error **errp)
         return;
     }
 
+    /* Connect the presenter to the VCPU (required for CPU hotplug) */
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_cpu_connect(tctx, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+
     qemu_register_reset(xive_tctx_reset, dev);
 }
 
@@ -893,6 +902,10 @@ static void xive_source_reset(void *dev)
 
     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
+
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_source_reset(xsrc, &error_fatal);
+    }
 }
 
 static void xive_source_realize(DeviceState *dev, Error **errp)
@@ -926,9 +939,11 @@ static void xive_source_realize(DeviceState *dev, Error **errp)
     xsrc->status = g_malloc0(xsrc->nr_irqs);
     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
 
-    memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
-                          &xive_source_esb_ops, xsrc, "xive.esb",
-                          (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
+    if (!kvm_irqchip_in_kernel()) {
+        memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
+                              &xive_source_esb_ops, xsrc, "xive.esb",
+                              (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
+    }
 
     qemu_register_reset(xive_source_reset, dev);
 }
hw/ppc/spapr_irq.c +5 -1
index 4145079d7fa5..6e1c36dc62ca 100644
@@ -387,7 +387,11 @@ static void spapr_irq_set_irq_xive(void *opaque, int srcno, int val)
 {
     sPAPRMachineState *spapr = opaque;
 
-    xive_source_set_irq(&spapr->xive->source, srcno, val);
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_source_set_irq(&spapr->xive->source, srcno, val);
+    } else {
+        xive_source_set_irq(&spapr->xive->source, srcno, val);
+    }
 }
 
 static const char *spapr_irq_get_nodename_xive(sPAPRMachineState *spapr)
target/ppc/kvm.c +7 -0
index d01852fe3112..43e42e3c2af9 100644
@@ -85,6 +85,7 @@ static int cap_fixup_hcalls;
 static int cap_htm;             /* Hardware transactional memory support */
 static int cap_mmu_radix;
 static int cap_mmu_hash_v3;
+static int cap_xive;
 static int cap_resize_hpt;
 static int cap_ppc_pvr_compat;
 static int cap_ppc_safe_cache;
@@ -148,6 +149,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
+    cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
     kvmppc_get_cpu_characteristics(s);
     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
@@ -2388,6 +2390,11 @@ static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
     return 0;
 }
 
+bool kvmppc_has_cap_xive(void)
+{
+    return cap_xive;
+}
+
 static void kvmppc_get_cpu_characteristics(KVMState *s)
 {
     struct kvm_ppc_cpu_char c;
hw/intc/Makefile.objs +1 -0
index 301a8e972d91..23126c199178 100644
@@ -39,6 +39,7 @@ obj-$(CONFIG_XICS_SPAPR) += xics_spapr.o
 obj-$(CONFIG_XICS_KVM) += xics_kvm.o
 obj-$(CONFIG_XIVE) += xive.o
 obj-$(CONFIG_XIVE_SPAPR) += spapr_xive.o
+obj-$(CONFIG_XIVE_KVM) += spapr_xive_kvm.o
 obj-$(CONFIG_POWERNV) += xics_pnv.o
 obj-$(CONFIG_ALLWINNER_A10_PIC) += allwinner-a10-pic.o
 obj-$(CONFIG_S390_FLIC) += s390_flic.o
Signature · 1 line
2.20.1
#2 lore raw

Re: [Qemu-devel] [PATCH v2 01/13] spapr/xive: add KVM support

Quoted 173 lines · Cédric Le Goater

This introduces a set of helpers when KVM is in use, which create the KVM XIVE device, initialize the interrupt sources at a KVM level and connect the interrupt presenters to the vCPU.

They also handle the initialization of the TIMA and the source ESB memory regions of the controller. These have a different type under KVM. They are 'ram device' memory mappings, similarly to VFIO, exposed to the guest and the associated VMAs on the host are populated dynamically with the appropriate pages using a fault handler.

---
 default-configs/ppc64-softmmu.mak |   1 +
 include/hw/ppc/spapr_xive.h       |  10 ++
 include/hw/ppc/xive.h             |  13 ++
 target/ppc/kvm_ppc.h              |   6 +
 hw/intc/spapr_xive.c              |  48 +++++-
 hw/intc/spapr_xive_kvm.c          | 237 ++++++++++++++++++++++++++++++
 hw/intc/xive.c                    |  21 ++-
 hw/ppc/spapr_irq.c                |   6 +-
 target/ppc/kvm.c                  |   7 +
 hw/intc/Makefile.objs             |   1 +
 10 files changed, 340 insertions(+), 10 deletions(-)
 create mode 100644 hw/intc/spapr_xive_kvm.c
default-configs/ppc64-softmmu.mak +1 -0
index 7f34ad0528ed..c1bf5cd951f5 100644
@@ -18,6 +18,7 @@ CONFIG_XICS_SPAPR=$(CONFIG_PSERIES)
 CONFIG_XICS_KVM=$(call land,$(CONFIG_PSERIES),$(CONFIG_KVM))
 CONFIG_XIVE=$(CONFIG_PSERIES)
 CONFIG_XIVE_SPAPR=$(CONFIG_PSERIES)
+CONFIG_XIVE_KVM=$(call land,$(CONFIG_PSERIES),$(CONFIG_KVM))
 CONFIG_MEM_DEVICE=y
 CONFIG_DIMM=y
 CONFIG_SPAPR_RNG=y
include/hw/ppc/spapr_xive.h +10 -0
index 2d31f24e3bfe..ab6732b14a02 100644
@@ -38,6 +38,10 @@ typedef struct sPAPRXive {
     /* TIMA mapping address */
     hwaddr        tm_base;
     MemoryRegion  tm_mmio;
+
+    /* KVM support */
+    int           fd;
+    void          *tm_mmap;
 } sPAPRXive;
 
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
@@ -49,5 +53,11 @@ void spapr_dt_xive(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
                    uint32_t phandle);
 void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx);
 void spapr_xive_mmio_set_enabled(sPAPRXive *xive, bool enable);
+void spapr_xive_map_mmio(sPAPRXive *xive);
+
+/*
+ * KVM XIVE device helpers
+ */
+void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +13 -0
index 13a487527b11..061d43fea24d 100644
@@ -140,6 +140,7 @@
 #ifndef PPC_XIVE_H
 #define PPC_XIVE_H
 
+#include "sysemu/kvm.h"
 #include "hw/qdev-core.h"
 #include "hw/sysbus.h"
 #include "hw/ppc/xive_regs.h"
@@ -194,6 +195,9 @@ typedef struct XiveSource {
     uint32_t        esb_shift;
     MemoryRegion    esb_mmio;
 
+    /* KVM support */
+    void            *esb_mmap;
+
     XiveNotifier    *xive;
 } XiveSource;
 
@@ -419,4 +423,13 @@ static inline uint32_t xive_nvt_cam_line(uint8_t nvt_blk, uint32_t nvt_idx)
     return (nvt_blk << 19) | nvt_idx;
 }
 
+/*
+ * KVM XIVE device helpers
+ */
+
+void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp);
+void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
+void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
+void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
+
 #endif /* PPC_XIVE_H */
target/ppc/kvm_ppc.h +6 -0
index bdfaa4e70a83..d2159660f9f2 100644
@@ -59,6 +59,7 @@ bool kvmppc_has_cap_fixup_hcalls(void);
 bool kvmppc_has_cap_htm(void);
 bool kvmppc_has_cap_mmu_radix(void);
 bool kvmppc_has_cap_mmu_hash_v3(void);
+bool kvmppc_has_cap_xive(void);
 int kvmppc_get_cap_safe_cache(void);
 int kvmppc_get_cap_safe_bounds_check(void);
 int kvmppc_get_cap_safe_indirect_branch(void);
@@ -307,6 +308,11 @@ static inline bool kvmppc_has_cap_mmu_hash_v3(void)
     return false;
 }
 
+static inline bool kvmppc_has_cap_xive(void)
+{
+    return false;
+}
+
 static inline int kvmppc_get_cap_safe_cache(void)
 {
     return 0;
hw/intc/spapr_xive.c +19 -1
index 06e3c9fdbfeb..c24d649e3668 100644
@@ -173,7 +173,7 @@ void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon)
     }
 }
 
-static void spapr_xive_map_mmio(sPAPRXive *xive)
+void spapr_xive_map_mmio(sPAPRXive *xive)
 {
     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
@@ -251,6 +251,9 @@ static void spapr_xive_instance_init(Object *obj)
                       TYPE_XIVE_END_SOURCE);
     object_property_add_child(obj, "end_source", OBJECT(&xive->end_source),
                               NULL);
+
+    /* Not connected to the KVM XIVE device */
+    xive->fd = -1;
 }
 
 static void spapr_xive_realize(DeviceState *dev, Error **errp)
@@ -259,6 +262,7 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
     XiveSource *xsrc = &xive->source;
     XiveENDSource *end_xsrc = &xive->end_source;
     Error *local_err = NULL;
+    MachineState *machine = MACHINE(qdev_get_machine());
 
     if (!xive->nr_irqs) {
         error_setg(errp, "Number of interrupt needs to be greater 0");
@@ -305,6 +309,32 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
     xive->endt = g_new0(XiveEND, xive->nr_ends);
 
+    xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
+                           xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
+
+    qemu_register_reset(spapr_xive_reset, dev);
+
+    if (kvm_enabled() && machine_kernel_irqchip_allowed(machine)) {
+        kvmppc_xive_connect(xive, &local_err);
+        if (local_err && machine_kernel_irqchip_required(machine)) {
+            error_prepend(&local_err,
+                          "kernel_irqchip requested but unavailable: ");
+            error_propagate(errp, local_err);
+            return;
+        }
+

hw/intc/spapr_xive.c static void spapr_xive_realize(DeviceState *dev, Error **errp) · line 335

Quoted 10 lines
hw/intc/spapr_xive.c +10 -0
+        if (!local_err) {
+            return;
+        }
+
+        /*
+         * We failed to initialize the XIVE KVM device, fallback to
+         * emulated mode
+         */
+        error_prepend(&local_err, "kernel_irqchip allowed but unavailable: ");
+        error_report_err(local_err);

Since we can fall back this should probably just be warn_report_err(). Maybe not even that, for the case where the host kernel doesn't support KVM XIVE at all.

Quoted 385 lines
hw/intc/spapr_xive.c +13 -5
+    }
+
     /* TIMA initialization */
     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
                           "xive.tima", 4ull << TM_SHIFT);
@@ -316,11 +346,6 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
 
     /* Map all regions */
     spapr_xive_map_mmio(xive);
-
-    xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
-                           xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
-
-    qemu_register_reset(spapr_xive_reset, dev);
 }
 
 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
@@ -495,6 +520,17 @@ bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi)
     if (lsi) {
         xive_source_irq_set_lsi(xsrc, lisn);
     }
+
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_source_reset_one(xsrc, lisn, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return false;
+        }
+    }
+
     return true;
 }
 
hw/intc/spapr_xive_kvm.c new file +237 -0
new file mode 100644
index 000000000000..623fbf74f23e
@@ -0,0 +1,237 @@
+/*
+ * QEMU PowerPC sPAPR XIVE interrupt controller model
+ *
+ * Copyright (c) 2017-2019, IBM Corporation.
+ *
+ * This code is licensed under the GPL version 2 or later. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "qemu/error-report.h"
+#include "qapi/error.h"
+#include "target/ppc/cpu.h"
+#include "sysemu/cpus.h"
+#include "sysemu/kvm.h"
+#include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_xive.h"
+#include "hw/ppc/xive.h"
+#include "kvm_ppc.h"
+
+#include <sys/ioctl.h>
+
+/*
+ * Helpers for CPU hotplug
+ *
+ * TODO: make a common KVMEnabledCPU layer for XICS and XIVE
+ */
+typedef struct KVMEnabledCPU {
+    unsigned long vcpu_id;
+    QLIST_ENTRY(KVMEnabledCPU) node;
+} KVMEnabledCPU;
+
+static QLIST_HEAD(, KVMEnabledCPU)
+    kvm_enabled_cpus = QLIST_HEAD_INITIALIZER(&kvm_enabled_cpus);
+
+static bool kvm_cpu_is_enabled(CPUState *cs)
+{
+    KVMEnabledCPU *enabled_cpu;
+    unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
+
+    QLIST_FOREACH(enabled_cpu, &kvm_enabled_cpus, node) {
+        if (enabled_cpu->vcpu_id == vcpu_id) {
+            return true;
+        }
+    }
+    return false;
+}
+
+static void kvm_cpu_enable(CPUState *cs)
+{
+    KVMEnabledCPU *enabled_cpu;
+    unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
+
+    enabled_cpu = g_malloc(sizeof(*enabled_cpu));
+    enabled_cpu->vcpu_id = vcpu_id;
+    QLIST_INSERT_HEAD(&kvm_enabled_cpus, enabled_cpu, node);
+}
+
+/*
+ * XIVE Thread Interrupt Management context (KVM)
+ */
+
+void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
+{
+    sPAPRXive *xive = SPAPR_MACHINE(qdev_get_machine())->xive;
+    unsigned long vcpu_id;
+    int ret;
+
+    /* Check if CPU was hot unplugged and replugged. */
+    if (kvm_cpu_is_enabled(tctx->cs)) {
+        return;
+    }
+
+    vcpu_id = kvm_arch_vcpu_id(tctx->cs);
+
+    ret = kvm_vcpu_enable_cap(tctx->cs, KVM_CAP_PPC_IRQ_XIVE, 0, xive->fd,
+                              vcpu_id, 0);
+    if (ret < 0) {
+        error_setg(errp, "XIVE: unable to connect CPU%ld to KVM device: %s",
+                   vcpu_id, strerror(errno));
+        return;
+    }
+
+    kvm_cpu_enable(tctx->cs);
+}
+
+/*
+ * XIVE Interrupt Source (KVM)
+ */
+
+/*
+ * At reset, the interrupt sources are simply created and MASKED. We
+ * only need to inform the KVM XIVE device about their type: LSI or
+ * MSI.
+ */
+void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp)
+{
+    sPAPRXive *xive = SPAPR_XIVE(xsrc->xive);
+    uint64_t state = 0;
+
+    if (xive_source_irq_is_lsi(xsrc, srcno)) {
+        state |= KVM_XIVE_LEVEL_SENSITIVE;
+        if (xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
+            state |= KVM_XIVE_LEVEL_ASSERTED;
+        }
+    }
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SOURCE, srcno, &state,
+                      true, errp);
+}
+
+void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
+{
+    int i;
+
+    for (i = 0; i < xsrc->nr_irqs; i++) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_source_reset_one(xsrc, i, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+}
+
+void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
+{
+    XiveSource *xsrc = opaque;
+    struct kvm_irq_level args;
+    int rc;
+
+    args.irq = srcno;
+    if (!xive_source_irq_is_lsi(xsrc, srcno)) {
+        if (!val) {
+            return;
+        }
+        args.level = KVM_INTERRUPT_SET;
+    } else {
+        if (val) {
+            xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
+            args.level = KVM_INTERRUPT_SET_LEVEL;
+        } else {
+            xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
+            args.level = KVM_INTERRUPT_UNSET;
+        }
+    }
+    rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
+    if (rc < 0) {
+        error_report("XIVE: kvm_irq_line() failed : %s", strerror(errno));
+    }
+}
+
+/*
+ * sPAPR XIVE interrupt controller (KVM)
+ */
+
+static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
+                              Error **errp)
+{
+    void *addr;
+    uint32_t page_shift = 16; /* TODO: fix page_shift */
+
+    addr = mmap(NULL, len, PROT_WRITE | PROT_READ, MAP_SHARED, xive->fd,
+                pgoff << page_shift);
+    if (addr == MAP_FAILED) {
+        error_setg_errno(errp, errno, "XIVE: unable to set memory mapping");
+        return NULL;
+    }
+
+    return addr;
+}
+
+/*
+ * All the XIVE memory regions are now backed by mappings from the KVM
+ * XIVE device.
+ */
+void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
+{
+    XiveSource *xsrc = &xive->source;
+    XiveENDSource *end_xsrc = &xive->end_source;
+    Error *local_err = NULL;
+    size_t esb_len = (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
+    size_t tima_len = 4ull << TM_SHIFT;
+
+    if (!kvmppc_has_cap_xive()) {
+        error_setg(errp, "IRQ_XIVE capability must be present for KVM");
+        return;
+    }
+
+    /* First, create the KVM XIVE device */
+    xive->fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_XIVE, false);
+    if (xive->fd < 0) {
+        error_setg_errno(errp, -xive->fd, "XIVE: error creating KVM device");
+        return;
+    }
+
+    /*
+     * 1. Source ESB pages - KVM mapping
+     */
+    xsrc->esb_mmap = kvmppc_xive_mmap(xive, KVM_XIVE_ESB_PAGE_OFFSET, esb_len,
+                                      &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
+    memory_region_init_ram_device_ptr(&xsrc->esb_mmio, OBJECT(xsrc),
+                                      "xive.esb", esb_len, xsrc->esb_mmap);
+    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
+
+    /*
+     * 2. END ESB pages (No KVM support yet)
+     */
+    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
+
+    /*
+     * 3. TIMA pages - KVM mapping
+     */
+    xive->tm_mmap = kvmppc_xive_mmap(xive, KVM_XIVE_TIMA_PAGE_OFFSET, tima_len,
+                                     &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+    memory_region_init_ram_device_ptr(&xive->tm_mmio, OBJECT(xive),
+                                      "xive.tima", tima_len, xive->tm_mmap);
+    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
+
+    kvm_kernel_irqchip = true;
+    kvm_msi_via_irqfd_allowed = true;
+    kvm_gsi_direct_mapping = true;
+
+    /* Map all regions */
+    spapr_xive_map_mmio(xive);
+}
hw/intc/xive.c +18 -3
index daa7badc8492..0284b5803551 100644
@@ -491,6 +491,15 @@ static void xive_tctx_realize(DeviceState *dev, Error **errp)
         return;
     }
 
+    /* Connect the presenter to the VCPU (required for CPU hotplug) */
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_cpu_connect(tctx, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+
     qemu_register_reset(xive_tctx_reset, dev);
 }
 
@@ -893,6 +902,10 @@ static void xive_source_reset(void *dev)
 
     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
+
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_source_reset(xsrc, &error_fatal);
+    }
 }
 
 static void xive_source_realize(DeviceState *dev, Error **errp)
@@ -926,9 +939,11 @@ static void xive_source_realize(DeviceState *dev, Error **errp)
     xsrc->status = g_malloc0(xsrc->nr_irqs);
     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
 
-    memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
-                          &xive_source_esb_ops, xsrc, "xive.esb",
-                          (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
+    if (!kvm_irqchip_in_kernel()) {
+        memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
+                              &xive_source_esb_ops, xsrc, "xive.esb",
+                              (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
+    }
 
     qemu_register_reset(xive_source_reset, dev);
 }
hw/ppc/spapr_irq.c +5 -1
index 4145079d7fa5..6e1c36dc62ca 100644
@@ -387,7 +387,11 @@ static void spapr_irq_set_irq_xive(void *opaque, int srcno, int val)
 {
     sPAPRMachineState *spapr = opaque;
 
-    xive_source_set_irq(&spapr->xive->source, srcno, val);
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_source_set_irq(&spapr->xive->source, srcno, val);
+    } else {
+        xive_source_set_irq(&spapr->xive->source, srcno, val);
+    }
 }
 
 static const char *spapr_irq_get_nodename_xive(sPAPRMachineState *spapr)
target/ppc/kvm.c +7 -0
index d01852fe3112..43e42e3c2af9 100644
@@ -85,6 +85,7 @@ static int cap_fixup_hcalls;
 static int cap_htm;             /* Hardware transactional memory support */
 static int cap_mmu_radix;
 static int cap_mmu_hash_v3;
+static int cap_xive;
 static int cap_resize_hpt;
 static int cap_ppc_pvr_compat;
 static int cap_ppc_safe_cache;
@@ -148,6 +149,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
+    cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
     kvmppc_get_cpu_characteristics(s);
     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
@@ -2388,6 +2390,11 @@ static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
     return 0;
 }
 
+bool kvmppc_has_cap_xive(void)
+{
+    return cap_xive;
+}
+
 static void kvmppc_get_cpu_characteristics(KVMState *s)
 {
     struct kvm_ppc_cpu_char c;
hw/intc/Makefile.objs +1 -0
index 301a8e972d91..23126c199178 100644
@@ -39,6 +39,7 @@ obj-$(CONFIG_XICS_SPAPR) += xics_spapr.o
 obj-$(CONFIG_XICS_KVM) += xics_kvm.o
 obj-$(CONFIG_XIVE) += xive.o
 obj-$(CONFIG_XIVE_SPAPR) += spapr_xive.o
+obj-$(CONFIG_XIVE_KVM) += spapr_xive_kvm.o
 obj-$(CONFIG_POWERNV) += xics_pnv.o
 obj-$(CONFIG_ALLWINNER_A10_PIC) += allwinner-a10-pic.o
 obj-$(CONFIG_S390_FLIC) += s390_flic.o
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
#3 lore raw

Re: [Qemu-devel] [PATCH v2 01/13] spapr/xive: add KVM support

Quoted 180 lines · David Gibson
Quoted 179 lines · Cédric Le Goater

This introduces a set of helpers when KVM is in use, which create the KVM XIVE device, initialize the interrupt sources at a KVM level and connect the interrupt presenters to the vCPU.

They also handle the initialization of the TIMA and the source ESB memory regions of the controller. These have a different type under KVM. They are 'ram device' memory mappings, similarly to VFIO, exposed to the guest and the associated VMAs on the host are populated dynamically with the appropriate pages using a fault handler.

---
 default-configs/ppc64-softmmu.mak |   1 +
 include/hw/ppc/spapr_xive.h       |  10 ++
 include/hw/ppc/xive.h             |  13 ++
 target/ppc/kvm_ppc.h              |   6 +
 hw/intc/spapr_xive.c              |  48 +++++-
 hw/intc/spapr_xive_kvm.c          | 237 ++++++++++++++++++++++++++++++
 hw/intc/xive.c                    |  21 ++-
 hw/ppc/spapr_irq.c                |   6 +-
 target/ppc/kvm.c                  |   7 +
 hw/intc/Makefile.objs             |   1 +
 10 files changed, 340 insertions(+), 10 deletions(-)
 create mode 100644 hw/intc/spapr_xive_kvm.c
default-configs/ppc64-softmmu.mak +1 -0
index 7f34ad0528ed..c1bf5cd951f5 100644
@@ -18,6 +18,7 @@ CONFIG_XICS_SPAPR=$(CONFIG_PSERIES)
 CONFIG_XICS_KVM=$(call land,$(CONFIG_PSERIES),$(CONFIG_KVM))
 CONFIG_XIVE=$(CONFIG_PSERIES)
 CONFIG_XIVE_SPAPR=$(CONFIG_PSERIES)
+CONFIG_XIVE_KVM=$(call land,$(CONFIG_PSERIES),$(CONFIG_KVM))
 CONFIG_MEM_DEVICE=y
 CONFIG_DIMM=y
 CONFIG_SPAPR_RNG=y
include/hw/ppc/spapr_xive.h +10 -0
index 2d31f24e3bfe..ab6732b14a02 100644
@@ -38,6 +38,10 @@ typedef struct sPAPRXive {
     /* TIMA mapping address */
     hwaddr        tm_base;
     MemoryRegion  tm_mmio;
+
+    /* KVM support */
+    int           fd;
+    void          *tm_mmap;
 } sPAPRXive;
 
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
@@ -49,5 +53,11 @@ void spapr_dt_xive(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
                    uint32_t phandle);
 void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx);
 void spapr_xive_mmio_set_enabled(sPAPRXive *xive, bool enable);
+void spapr_xive_map_mmio(sPAPRXive *xive);
+
+/*
+ * KVM XIVE device helpers
+ */
+void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +13 -0
index 13a487527b11..061d43fea24d 100644
@@ -140,6 +140,7 @@
 #ifndef PPC_XIVE_H
 #define PPC_XIVE_H
 
+#include "sysemu/kvm.h"
 #include "hw/qdev-core.h"
 #include "hw/sysbus.h"
 #include "hw/ppc/xive_regs.h"
@@ -194,6 +195,9 @@ typedef struct XiveSource {
     uint32_t        esb_shift;
     MemoryRegion    esb_mmio;
 
+    /* KVM support */
+    void            *esb_mmap;
+
     XiveNotifier    *xive;
 } XiveSource;
 
@@ -419,4 +423,13 @@ static inline uint32_t xive_nvt_cam_line(uint8_t nvt_blk, uint32_t nvt_idx)
     return (nvt_blk << 19) | nvt_idx;
 }
 
+/*
+ * KVM XIVE device helpers
+ */
+
+void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp);
+void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
+void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
+void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
+
 #endif /* PPC_XIVE_H */
target/ppc/kvm_ppc.h +6 -0
index bdfaa4e70a83..d2159660f9f2 100644
@@ -59,6 +59,7 @@ bool kvmppc_has_cap_fixup_hcalls(void);
 bool kvmppc_has_cap_htm(void);
 bool kvmppc_has_cap_mmu_radix(void);
 bool kvmppc_has_cap_mmu_hash_v3(void);
+bool kvmppc_has_cap_xive(void);
 int kvmppc_get_cap_safe_cache(void);
 int kvmppc_get_cap_safe_bounds_check(void);
 int kvmppc_get_cap_safe_indirect_branch(void);
@@ -307,6 +308,11 @@ static inline bool kvmppc_has_cap_mmu_hash_v3(void)
     return false;
 }
 
+static inline bool kvmppc_has_cap_xive(void)
+{
+    return false;
+}
+
 static inline int kvmppc_get_cap_safe_cache(void)
 {
     return 0;
hw/intc/spapr_xive.c +25 -1
index 06e3c9fdbfeb..c24d649e3668 100644
@@ -173,7 +173,7 @@ void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon)
     }
 }
 
-static void spapr_xive_map_mmio(sPAPRXive *xive)
+void spapr_xive_map_mmio(sPAPRXive *xive)
 {
     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 0, xive->vc_base);
     sysbus_mmio_map(SYS_BUS_DEVICE(xive), 1, xive->end_base);
@@ -251,6 +251,9 @@ static void spapr_xive_instance_init(Object *obj)
                       TYPE_XIVE_END_SOURCE);
     object_property_add_child(obj, "end_source", OBJECT(&xive->end_source),
                               NULL);
+
+    /* Not connected to the KVM XIVE device */
+    xive->fd = -1;
 }
 
 static void spapr_xive_realize(DeviceState *dev, Error **errp)
@@ -259,6 +262,7 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
     XiveSource *xsrc = &xive->source;
     XiveENDSource *end_xsrc = &xive->end_source;
     Error *local_err = NULL;
+    MachineState *machine = MACHINE(qdev_get_machine());
 
     if (!xive->nr_irqs) {
         error_setg(errp, "Number of interrupt needs to be greater 0");
@@ -305,6 +309,32 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
     xive->eat = g_new0(XiveEAS, xive->nr_irqs);
     xive->endt = g_new0(XiveEND, xive->nr_ends);
 
+    xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
+                           xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
+
+    qemu_register_reset(spapr_xive_reset, dev);
+
+    if (kvm_enabled() && machine_kernel_irqchip_allowed(machine)) {
+        kvmppc_xive_connect(xive, &local_err);
+        if (local_err && machine_kernel_irqchip_required(machine)) {
+            error_prepend(&local_err,
+                          "kernel_irqchip requested but unavailable: ");
+            error_propagate(errp, local_err);
+            return;
+        }
+
+        if (!local_err) {
+            return;
+        }
+
+        /*
+         * We failed to initialize the XIVE KVM device, fallback to
Quoted 8 lines

hw/intc/spapr_xive.c static void spapr_xive_realize(DeviceState *dev, Error **errp) · line 335

Quoted 4 lines
hw/intc/spapr_xive.c +4 -0
+         * emulated mode
+         */
+        error_prepend(&local_err, "kernel_irqchip allowed but unavailable: ");
+        error_report_err(local_err);

Since we can fall back this should probably just be warn_report_err(). Maybe not even that, for the case where the host kernel doesn't support KVM XIVE at all.

Let's use a warn_report_err().

Thanks,

C.

Quoted 386 lines
Quoted 385 lines
hw/intc/spapr_xive.c +13 -5
+    }
+
     /* TIMA initialization */
     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
                           "xive.tima", 4ull << TM_SHIFT);
@@ -316,11 +346,6 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
 
     /* Map all regions */
     spapr_xive_map_mmio(xive);
-
-    xive->nodename = g_strdup_printf("interrupt-controller@%" PRIx64,
-                           xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
-
-    qemu_register_reset(spapr_xive_reset, dev);
 }
 
 static int spapr_xive_get_eas(XiveRouter *xrtr, uint8_t eas_blk,
@@ -495,6 +520,17 @@ bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi)
     if (lsi) {
         xive_source_irq_set_lsi(xsrc, lisn);
     }
+
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_source_reset_one(xsrc, lisn, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return false;
+        }
+    }
+
     return true;
 }
 
hw/intc/spapr_xive_kvm.c new file +237 -0
new file mode 100644
index 000000000000..623fbf74f23e
@@ -0,0 +1,237 @@
+/*
+ * QEMU PowerPC sPAPR XIVE interrupt controller model
+ *
+ * Copyright (c) 2017-2019, IBM Corporation.
+ *
+ * This code is licensed under the GPL version 2 or later. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "qemu/error-report.h"
+#include "qapi/error.h"
+#include "target/ppc/cpu.h"
+#include "sysemu/cpus.h"
+#include "sysemu/kvm.h"
+#include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_xive.h"
+#include "hw/ppc/xive.h"
+#include "kvm_ppc.h"
+
+#include <sys/ioctl.h>
+
+/*
+ * Helpers for CPU hotplug
+ *
+ * TODO: make a common KVMEnabledCPU layer for XICS and XIVE
+ */
+typedef struct KVMEnabledCPU {
+    unsigned long vcpu_id;
+    QLIST_ENTRY(KVMEnabledCPU) node;
+} KVMEnabledCPU;
+
+static QLIST_HEAD(, KVMEnabledCPU)
+    kvm_enabled_cpus = QLIST_HEAD_INITIALIZER(&kvm_enabled_cpus);
+
+static bool kvm_cpu_is_enabled(CPUState *cs)
+{
+    KVMEnabledCPU *enabled_cpu;
+    unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
+
+    QLIST_FOREACH(enabled_cpu, &kvm_enabled_cpus, node) {
+        if (enabled_cpu->vcpu_id == vcpu_id) {
+            return true;
+        }
+    }
+    return false;
+}
+
+static void kvm_cpu_enable(CPUState *cs)
+{
+    KVMEnabledCPU *enabled_cpu;
+    unsigned long vcpu_id = kvm_arch_vcpu_id(cs);
+
+    enabled_cpu = g_malloc(sizeof(*enabled_cpu));
+    enabled_cpu->vcpu_id = vcpu_id;
+    QLIST_INSERT_HEAD(&kvm_enabled_cpus, enabled_cpu, node);
+}
+
+/*
+ * XIVE Thread Interrupt Management context (KVM)
+ */
+
+void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
+{
+    sPAPRXive *xive = SPAPR_MACHINE(qdev_get_machine())->xive;
+    unsigned long vcpu_id;
+    int ret;
+
+    /* Check if CPU was hot unplugged and replugged. */
+    if (kvm_cpu_is_enabled(tctx->cs)) {
+        return;
+    }
+
+    vcpu_id = kvm_arch_vcpu_id(tctx->cs);
+
+    ret = kvm_vcpu_enable_cap(tctx->cs, KVM_CAP_PPC_IRQ_XIVE, 0, xive->fd,
+                              vcpu_id, 0);
+    if (ret < 0) {
+        error_setg(errp, "XIVE: unable to connect CPU%ld to KVM device: %s",
+                   vcpu_id, strerror(errno));
+        return;
+    }
+
+    kvm_cpu_enable(tctx->cs);
+}
+
+/*
+ * XIVE Interrupt Source (KVM)
+ */
+
+/*
+ * At reset, the interrupt sources are simply created and MASKED. We
+ * only need to inform the KVM XIVE device about their type: LSI or
+ * MSI.
+ */
+void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp)
+{
+    sPAPRXive *xive = SPAPR_XIVE(xsrc->xive);
+    uint64_t state = 0;
+
+    if (xive_source_irq_is_lsi(xsrc, srcno)) {
+        state |= KVM_XIVE_LEVEL_SENSITIVE;
+        if (xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
+            state |= KVM_XIVE_LEVEL_ASSERTED;
+        }
+    }
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SOURCE, srcno, &state,
+                      true, errp);
+}
+
+void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
+{
+    int i;
+
+    for (i = 0; i < xsrc->nr_irqs; i++) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_source_reset_one(xsrc, i, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+}
+
+void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
+{
+    XiveSource *xsrc = opaque;
+    struct kvm_irq_level args;
+    int rc;
+
+    args.irq = srcno;
+    if (!xive_source_irq_is_lsi(xsrc, srcno)) {
+        if (!val) {
+            return;
+        }
+        args.level = KVM_INTERRUPT_SET;
+    } else {
+        if (val) {
+            xsrc->status[srcno] |= XIVE_STATUS_ASSERTED;
+            args.level = KVM_INTERRUPT_SET_LEVEL;
+        } else {
+            xsrc->status[srcno] &= ~XIVE_STATUS_ASSERTED;
+            args.level = KVM_INTERRUPT_UNSET;
+        }
+    }
+    rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
+    if (rc < 0) {
+        error_report("XIVE: kvm_irq_line() failed : %s", strerror(errno));
+    }
+}
+
+/*
+ * sPAPR XIVE interrupt controller (KVM)
+ */
+
+static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
+                              Error **errp)
+{
+    void *addr;
+    uint32_t page_shift = 16; /* TODO: fix page_shift */
+
+    addr = mmap(NULL, len, PROT_WRITE | PROT_READ, MAP_SHARED, xive->fd,
+                pgoff << page_shift);
+    if (addr == MAP_FAILED) {
+        error_setg_errno(errp, errno, "XIVE: unable to set memory mapping");
+        return NULL;
+    }
+
+    return addr;
+}
+
+/*
+ * All the XIVE memory regions are now backed by mappings from the KVM
+ * XIVE device.
+ */
+void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
+{
+    XiveSource *xsrc = &xive->source;
+    XiveENDSource *end_xsrc = &xive->end_source;
+    Error *local_err = NULL;
+    size_t esb_len = (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
+    size_t tima_len = 4ull << TM_SHIFT;
+
+    if (!kvmppc_has_cap_xive()) {
+        error_setg(errp, "IRQ_XIVE capability must be present for KVM");
+        return;
+    }
+
+    /* First, create the KVM XIVE device */
+    xive->fd = kvm_create_device(kvm_state, KVM_DEV_TYPE_XIVE, false);
+    if (xive->fd < 0) {
+        error_setg_errno(errp, -xive->fd, "XIVE: error creating KVM device");
+        return;
+    }
+
+    /*
+     * 1. Source ESB pages - KVM mapping
+     */
+    xsrc->esb_mmap = kvmppc_xive_mmap(xive, KVM_XIVE_ESB_PAGE_OFFSET, esb_len,
+                                      &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
+    memory_region_init_ram_device_ptr(&xsrc->esb_mmio, OBJECT(xsrc),
+                                      "xive.esb", esb_len, xsrc->esb_mmap);
+    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xsrc->esb_mmio);
+
+    /*
+     * 2. END ESB pages (No KVM support yet)
+     */
+    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &end_xsrc->esb_mmio);
+
+    /*
+     * 3. TIMA pages - KVM mapping
+     */
+    xive->tm_mmap = kvmppc_xive_mmap(xive, KVM_XIVE_TIMA_PAGE_OFFSET, tima_len,
+                                     &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+    memory_region_init_ram_device_ptr(&xive->tm_mmio, OBJECT(xive),
+                                      "xive.tima", tima_len, xive->tm_mmap);
+    sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
+
+    kvm_kernel_irqchip = true;
+    kvm_msi_via_irqfd_allowed = true;
+    kvm_gsi_direct_mapping = true;
+
+    /* Map all regions */
+    spapr_xive_map_mmio(xive);
+}
hw/intc/xive.c +18 -3
index daa7badc8492..0284b5803551 100644
@@ -491,6 +491,15 @@ static void xive_tctx_realize(DeviceState *dev, Error **errp)
         return;
     }
 
+    /* Connect the presenter to the VCPU (required for CPU hotplug) */
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_cpu_connect(tctx, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+
     qemu_register_reset(xive_tctx_reset, dev);
 }
 
@@ -893,6 +902,10 @@ static void xive_source_reset(void *dev)
 
     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
+
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_source_reset(xsrc, &error_fatal);
+    }
 }
 
 static void xive_source_realize(DeviceState *dev, Error **errp)
@@ -926,9 +939,11 @@ static void xive_source_realize(DeviceState *dev, Error **errp)
     xsrc->status = g_malloc0(xsrc->nr_irqs);
     xsrc->lsi_map = bitmap_new(xsrc->nr_irqs);
 
-    memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
-                          &xive_source_esb_ops, xsrc, "xive.esb",
-                          (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
+    if (!kvm_irqchip_in_kernel()) {
+        memory_region_init_io(&xsrc->esb_mmio, OBJECT(xsrc),
+                              &xive_source_esb_ops, xsrc, "xive.esb",
+                              (1ull << xsrc->esb_shift) * xsrc->nr_irqs);
+    }
 
     qemu_register_reset(xive_source_reset, dev);
 }
hw/ppc/spapr_irq.c +5 -1
index 4145079d7fa5..6e1c36dc62ca 100644
@@ -387,7 +387,11 @@ static void spapr_irq_set_irq_xive(void *opaque, int srcno, int val)
 {
     sPAPRMachineState *spapr = opaque;
 
-    xive_source_set_irq(&spapr->xive->source, srcno, val);
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_source_set_irq(&spapr->xive->source, srcno, val);
+    } else {
+        xive_source_set_irq(&spapr->xive->source, srcno, val);
+    }
 }
 
 static const char *spapr_irq_get_nodename_xive(sPAPRMachineState *spapr)
target/ppc/kvm.c +7 -0
index d01852fe3112..43e42e3c2af9 100644
@@ -85,6 +85,7 @@ static int cap_fixup_hcalls;
 static int cap_htm;             /* Hardware transactional memory support */
 static int cap_mmu_radix;
 static int cap_mmu_hash_v3;
+static int cap_xive;
 static int cap_resize_hpt;
 static int cap_ppc_pvr_compat;
 static int cap_ppc_safe_cache;
@@ -148,6 +149,7 @@ int kvm_arch_init(MachineState *ms, KVMState *s)
     cap_htm = kvm_vm_check_extension(s, KVM_CAP_PPC_HTM);
     cap_mmu_radix = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_RADIX);
     cap_mmu_hash_v3 = kvm_vm_check_extension(s, KVM_CAP_PPC_MMU_HASH_V3);
+    cap_xive = kvm_vm_check_extension(s, KVM_CAP_PPC_IRQ_XIVE);
     cap_resize_hpt = kvm_vm_check_extension(s, KVM_CAP_SPAPR_RESIZE_HPT);
     kvmppc_get_cpu_characteristics(s);
     cap_ppc_nested_kvm_hv = kvm_vm_check_extension(s, KVM_CAP_PPC_NESTED_HV);
@@ -2388,6 +2390,11 @@ static int parse_cap_ppc_safe_indirect_branch(struct kvm_ppc_cpu_char c)
     return 0;
 }
 
+bool kvmppc_has_cap_xive(void)
+{
+    return cap_xive;
+}
+
 static void kvmppc_get_cpu_characteristics(KVMState *s)
 {
     struct kvm_ppc_cpu_char c;
hw/intc/Makefile.objs +1 -0
index 301a8e972d91..23126c199178 100644
@@ -39,6 +39,7 @@ obj-$(CONFIG_XICS_SPAPR) += xics_spapr.o
 obj-$(CONFIG_XICS_KVM) += xics_kvm.o
 obj-$(CONFIG_XIVE) += xive.o
 obj-$(CONFIG_XIVE_SPAPR) += spapr_xive.o
+obj-$(CONFIG_XIVE_KVM) += spapr_xive_kvm.o
 obj-$(CONFIG_POWERNV) += xics_pnv.o
 obj-$(CONFIG_ALLWINNER_A10_PIC) += allwinner-a10-pic.o
 obj-$(CONFIG_S390_FLIC) += s390_flic.o
#4 lore raw

PATCH v2 2/13 spapr/xive: add hcall support when under KVM

In reply to Cédric Le Goater

XIVE hcalls are all redirected to QEMU as none are on a fast path. When necessary, QEMU invokes KVM through specific ioctls to perform host operations. QEMU should have done the necessary checks before calling KVM and, in case of failure, H_HARDWARE is simply returned.

H_INT_ESB is a special case that could have been handled under KVM but the impact on performance was low when under QEMU. Here are some figures :

kernel irqchip OFF ON
H_INT_ESB KVM QEMU

rtl8139 (LSI ) 1.19 1.24 1.23 Gbits/sec
virtio 31.80 42.30 -- Gbits/sec

---
 include/hw/ppc/spapr_xive.h |  15 +++
 hw/intc/spapr_xive.c        |  87 +++++++++++++++--
 hw/intc/spapr_xive_kvm.c    | 184 ++++++++++++++++++++++++++++++++++++
 3 files changed, 278 insertions(+), 8 deletions(-)
include/hw/ppc/spapr_xive.h +15 -0
index ab6732b14a02..749c6cbc2c56 100644
@@ -55,9 +55,24 @@ void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx);
 void spapr_xive_mmio_set_enabled(sPAPRXive *xive, bool enable);
 void spapr_xive_map_mmio(sPAPRXive *xive);
 
+int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
+                             uint32_t *out_server, uint8_t *out_prio);
+
 /*
  * KVM XIVE device helpers
  */
 void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_reset(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
+                                   Error **errp);
+void kvmppc_xive_sync_source(sPAPRXive *xive, uint32_t lisn, Error **errp);
+uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write);
+void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                 uint32_t end_idx, XiveEND *end,
+                                 Error **errp);
+void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                 uint32_t end_idx, XiveEND *end,
+                                 Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
hw/intc/spapr_xive.c +79 -8

131 lines not on this page · raw message

hw/intc/spapr_xive_kvm.c +184 -0

202 lines not on this page · raw message

Signature · 1 line
2.20.1
#5 lore raw

Re: [Qemu-devel] [PATCH v2 02/13] spapr/xive: add hcall support when under KVM

Quoted 56 lines · Cédric Le Goater

XIVE hcalls are all redirected to QEMU as none are on a fast path. When necessary, QEMU invokes KVM through specific ioctls to perform host operations. QEMU should have done the necessary checks before calling KVM and, in case of failure, H_HARDWARE is simply returned.

H_INT_ESB is a special case that could have been handled under KVM but the impact on performance was low when under QEMU. Here are some figures :

kernel irqchip OFF ON
H_INT_ESB KVM QEMU

rtl8139 (LSI ) 1.19 1.24 1.23 Gbits/sec
virtio 31.80 42.30 -- Gbits/sec

---
 include/hw/ppc/spapr_xive.h |  15 +++
 hw/intc/spapr_xive.c        |  87 +++++++++++++++--
 hw/intc/spapr_xive_kvm.c    | 184 ++++++++++++++++++++++++++++++++++++
 3 files changed, 278 insertions(+), 8 deletions(-)
include/hw/ppc/spapr_xive.h +15 -0
index ab6732b14a02..749c6cbc2c56 100644
@@ -55,9 +55,24 @@ void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx);
 void spapr_xive_mmio_set_enabled(sPAPRXive *xive, bool enable);
 void spapr_xive_map_mmio(sPAPRXive *xive);
 
+int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
+                             uint32_t *out_server, uint8_t *out_prio);
+
 /*
  * KVM XIVE device helpers
  */
 void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_reset(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
+                                   Error **errp);
+void kvmppc_xive_sync_source(sPAPRXive *xive, uint32_t lisn, Error **errp);
+uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write);
+void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                 uint32_t end_idx, XiveEND *end,
+                                 Error **errp);
+void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                 uint32_t end_idx, XiveEND *end,
+                                 Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
hw/intc/spapr_xive.c +0 -0
index c24d649e3668..3db24391e31c 100644

hw/intc/spapr_xive.c static int spapr_xive_target_to_nvt(uint32_t target, · line 91

Quoted 6 lines
hw/intc/spapr_xive.c +3 -0
@@ -86,6 +86,19 @@ static int spapr_xive_target_to_nvt(uint32_t target,
  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
  * priorities per CPU
  */
+int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
+                             uint32_t *out_server, uint8_t *out_prio)
+{

Since you don't support irq blocks as yet, should this error out rather than ignoring if end_blk != 0?

Quoted 143 lines
hw/intc/spapr_xive.c +76 -8
+    if (out_server) {
+        *out_server = end_idx >> 3;
+    }
+
+    if (out_prio) {
+        *out_prio = end_idx & 0x7;
+    }
+    return 0;
+}
+
 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
 {
@@ -792,6 +805,16 @@ static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
     }
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
+
 out:
     xive->eat[lisn] = new_eas;
     return H_SUCCESS;
@@ -1097,6 +1120,16 @@ static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
      */
 
 out:
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
+
     /* Update END */
     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
     return H_SUCCESS;
@@ -1189,6 +1222,16 @@ static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
         args[2] = 0;
     }
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
+
     /* TODO: do we need any locking on the END ? */
     if (flags & SPAPR_XIVE_END_DEBUG) {
         /* Load the event queue generation number into the return flags */
@@ -1341,15 +1384,20 @@ static target_ulong h_int_esb(PowerPCCPU *cpu,
         return H_P3;
     }
 
-    mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
+    if (kvm_irqchip_in_kernel()) {
+        args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
+                                     flags & SPAPR_XIVE_ESB_STORE);
+    } else {
+        mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
 
-    if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
-                      (flags & SPAPR_XIVE_ESB_STORE))) {
-        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
-                      HWADDR_PRIx "\n", mmio_addr);
-        return H_HARDWARE;
+        if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
+                          (flags & SPAPR_XIVE_ESB_STORE))) {
+            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
+                          HWADDR_PRIx "\n", mmio_addr);
+            return H_HARDWARE;
+        }
+        args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
     }
-    args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
     return H_SUCCESS;
 }
 
@@ -1406,7 +1454,20 @@ static target_ulong h_int_sync(PowerPCCPU *cpu,
      * This is not needed when running the emulation under QEMU
      */
 
-    /* This is not real hardware. Nothing to be done */
+    /*
+     * This is not real hardware. Nothing to be done unless when
+     * under KVM
+     */
+
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_sync_source(xive, lisn, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
     return H_SUCCESS;
 }
 
@@ -1441,6 +1502,16 @@ static target_ulong h_int_reset(PowerPCCPU *cpu,
     }
 
     device_reset(DEVICE(xive));
+
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_reset(xive, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
     return H_SUCCESS;
 }
 
hw/intc/spapr_xive_kvm.c +4 -0
index 623fbf74f23e..6b50451b4f85 100644
@@ -89,6 +89,52 @@ void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
  * XIVE Interrupt Source (KVM)
  */
 
+void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
+                                   Error **errp)
+{
+    uint32_t end_idx;

hw/intc/spapr_xive_kvm.c void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp) · line 105

Quoted 10 lines
hw/intc/spapr_xive_kvm.c +10 -0
+    uint32_t end_blk;
+    uint32_t eisn;
+    uint8_t priority;
+    uint32_t server;
+    uint64_t kvm_src;
+    Error *local_err = NULL;
+
+    /*
+     * No need to set a MASKED source, this is the default state after
+     * reset.

I don't quite follow this comment, why is there no need to call a MASKED source?

Quoted 38 lines
hw/intc/spapr_xive_kvm.c +32 -0
+     */
+    if (!xive_eas_is_valid(eas) || xive_eas_is_masked(eas)) {
+        return;
+    }
+
+    end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
+    end_blk = xive_get_field64(EAS_END_BLOCK, eas->w);
+    eisn = xive_get_field64(EAS_END_DATA, eas->w);
+
+    spapr_xive_end_to_target(end_blk, end_idx, &server, &priority);
+
+    kvm_src = priority << KVM_XIVE_SOURCE_PRIORITY_SHIFT &
+        KVM_XIVE_SOURCE_PRIORITY_MASK;
+    kvm_src |= server << KVM_XIVE_SOURCE_SERVER_SHIFT &
+        KVM_XIVE_SOURCE_SERVER_MASK;
+    kvm_src |= ((uint64_t)eisn << KVM_XIVE_SOURCE_EISN_SHIFT) &
+        KVM_XIVE_SOURCE_EISN_MASK;
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SOURCE_CONFIG, lisn,
+                      &kvm_src, true, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+}
+
+void kvmppc_xive_sync_source(sPAPRXive *xive, uint32_t lisn, Error **errp)
+{
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SOURCE_SYNC, lisn,
+                      NULL, true, errp);
+}
+
 /*
  * At reset, the interrupt sources are simply created and MASKED. We
  * only need to inform the KVM XIVE device about their type: LSI or
@@ -125,6 +171,64 @@ void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
     }
 }

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 182

Quoted 10 lines
hw/intc/spapr_xive_kvm.c +9 -0
 
+/*
+ * This is used to perform the magic loads on the ESB pages, described
+ * in xive.h.
+ */
+static uint64_t xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write)
+{
+    unsigned long addr = (unsigned long) xsrc->esb_mmap +
+        xive_source_esb_mgmt(xsrc, srcno) + offset;

Casting the esb_mmap into unsigned long then back to a pointer looks unnecessary. You should be able to do this with pointer arithmetic.

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 189

Quoted 6 lines
hw/intc/spapr_xive_kvm.c +6 -0
+    if (write) {
+        *((uint64_t *) addr) = data;
+        return -1;
+    } else {
+        return *((uint64_t *) addr);
+    }

Since this is always dealing with 64-bit values, couldn't you put the byteswaps in here rather than in all the callers?

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 195

Quoted 6 lines
hw/intc/spapr_xive_kvm.c +6 -0
+}
+
+static uint8_t xive_esb_read(XiveSource *xsrc, int srcno, uint32_t offset)
+{
+    /* Prevent the compiler from optimizing away the load */
+    volatile uint64_t value = xive_esb_rw(xsrc, srcno, offset, 0, 0);

Wouldn't the volatile magic be better inside xive_esb_rw()?

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 206

Quoted 10 lines
hw/intc/spapr_xive_kvm.c +10 -0
+    return be64_to_cpu(value) & 0x3;
+}
+
+static void xive_esb_trigger(XiveSource *xsrc, int srcno)
+{
+    unsigned long addr = (unsigned long) xsrc->esb_mmap +
+        xive_source_esb_page(xsrc, srcno);
+
+    *((uint64_t *) addr) = 0x0;
+}

Also.. aren't some of these register accesses likely to need memory barriers?

Quoted 33 lines
hw/intc/spapr_xive_kvm.c +26 -0
+
+uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write)
+{
+    if (write) {
+        return xive_esb_rw(xsrc, srcno, offset, data, 1);
+    }
+
+    /*
+     * Special Load EOI handling for LSI sources. Q bit is never set
+     * and the interrupt should be re-triggered if the level is still
+     * asserted.
+     */
+    if (xive_source_irq_is_lsi(xsrc, srcno) &&
+        offset == XIVE_ESB_LOAD_EOI) {
+        xive_esb_read(xsrc, srcno, XIVE_ESB_SET_PQ_00);
+        if (xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
+            xive_esb_trigger(xsrc, srcno);
+        }
+        return 0;
+    } else {
+        return xive_esb_rw(xsrc, srcno, offset, 0, 0);
+    }
+}
+
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 {
     XiveSource *xsrc = opaque;
@@ -155,6 +259,86 @@ void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 /*
  * sPAPR XIVE interrupt controller (KVM)
  */
+void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val) · line 272

Quoted 10 lines
hw/intc/spapr_xive_kvm.c +10 -0
+                                  uint32_t end_idx, XiveEND *end,
+                                  Error **errp)
+{
+    struct kvm_ppc_xive_eq kvm_eq = { 0 };
+    uint64_t kvm_eq_idx;
+    uint8_t priority;
+    uint32_t server;
+    Error *local_err = NULL;
+
+    if (!xive_end_is_valid(end)) {

This should set an error, shouldn't it?

Quoted 72 lines
hw/intc/spapr_xive_kvm.c +69 -0
+        return;
+    }
+
+    /* Encode the tuple (server, prio) as a KVM EQ index */
+    spapr_xive_end_to_target(end_blk, end_idx, &server, &priority);
+
+    kvm_eq_idx = priority << KVM_XIVE_EQ_PRIORITY_SHIFT &
+            KVM_XIVE_EQ_PRIORITY_MASK;
+    kvm_eq_idx |= server << KVM_XIVE_EQ_SERVER_SHIFT &
+        KVM_XIVE_EQ_SERVER_MASK;
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_EQ_CONFIG, kvm_eq_idx,
+                      &kvm_eq, false, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
+    /*
+     * The EQ index and toggle bit are updated by HW. These are the
+     * only fields we want to return.
+     */
+    end->w1 = xive_set_field32(END_W1_GENERATION, 0ul, kvm_eq.qtoggle) |
+        xive_set_field32(END_W1_PAGE_OFF, 0ul, kvm_eq.qindex);
+}
+
+void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                  uint32_t end_idx, XiveEND *end,
+                                  Error **errp)
+{
+    struct kvm_ppc_xive_eq kvm_eq = { 0 };
+    uint64_t kvm_eq_idx;
+    uint8_t priority;
+    uint32_t server;
+    Error *local_err = NULL;
+
+    if (!xive_end_is_valid(end)) {
+        return;
+    }
+
+    /* Build the KVM state from the local END structure */
+    kvm_eq.flags   = KVM_XIVE_EQ_FLAG_ALWAYS_NOTIFY;
+    kvm_eq.qsize   = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
+    kvm_eq.qpage   = (uint64_t) be32_to_cpu(end->w2 & 0x0fffffff) << 32 |
+        be32_to_cpu(end->w3);
+    kvm_eq.qtoggle = xive_get_field32(END_W1_GENERATION, end->w1);
+    kvm_eq.qindex  = xive_get_field32(END_W1_PAGE_OFF, end->w1);
+
+    /* Encode the tuple (server, prio) as a KVM EQ index */
+    spapr_xive_end_to_target(end_blk, end_idx, &server, &priority);
+
+    kvm_eq_idx = priority << KVM_XIVE_EQ_PRIORITY_SHIFT &
+            KVM_XIVE_EQ_PRIORITY_MASK;
+    kvm_eq_idx |= server << KVM_XIVE_EQ_SERVER_SHIFT &
+        KVM_XIVE_EQ_SERVER_MASK;
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_EQ_CONFIG, kvm_eq_idx,
+                      &kvm_eq, true, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+}
+
+void kvmppc_xive_reset(sPAPRXive *xive, Error **errp)
+{
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_CTRL, KVM_DEV_XIVE_RESET,
+                      NULL, true, errp);
+}
 
 static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
                               Error **errp)
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
#6 lore raw

Re: [Qemu-devel] [PATCH v2 02/13] spapr/xive: add hcall support when under KVM

Quoted 51 lines · David Gibson
Quoted 50 lines · Cédric Le Goater

XIVE hcalls are all redirected to QEMU as none are on a fast path. When necessary, QEMU invokes KVM through specific ioctls to perform host operations. QEMU should have done the necessary checks before calling KVM and, in case of failure, H_HARDWARE is simply returned.

H_INT_ESB is a special case that could have been handled under KVM but the impact on performance was low when under QEMU. Here are some figures :

kernel irqchip OFF ON
H_INT_ESB KVM QEMU

rtl8139 (LSI ) 1.19 1.24 1.23 Gbits/sec
virtio 31.80 42.30 -- Gbits/sec

---
 include/hw/ppc/spapr_xive.h |  15 +++
 hw/intc/spapr_xive.c        |  87 +++++++++++++++--
 hw/intc/spapr_xive_kvm.c    | 184 ++++++++++++++++++++++++++++++++++++
 3 files changed, 278 insertions(+), 8 deletions(-)
include/hw/ppc/spapr_xive.h +15 -0
index ab6732b14a02..749c6cbc2c56 100644
@@ -55,9 +55,24 @@ void spapr_xive_set_tctx_os_cam(XiveTCTX *tctx);
 void spapr_xive_mmio_set_enabled(sPAPRXive *xive, bool enable);
 void spapr_xive_map_mmio(sPAPRXive *xive);
 
+int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
+                             uint32_t *out_server, uint8_t *out_prio);
+
 /*
  * KVM XIVE device helpers
  */
 void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_reset(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
+                                   Error **errp);
+void kvmppc_xive_sync_source(sPAPRXive *xive, uint32_t lisn, Error **errp);
+uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write);
+void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                 uint32_t end_idx, XiveEND *end,
+                                 Error **errp);
+void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                 uint32_t end_idx, XiveEND *end,
+                                 Error **errp);
 
Quoted 15 lines

hw/intc/spapr_xive.c static int spapr_xive_target_to_nvt(uint32_t target, · line 91

Quoted 12 lines
include/hw/ppc/spapr_xive.h +0 -0
 #endif /* PPC_SPAPR_XIVE_H */
hw/intc/spapr_xive.c +3 -0
index c24d649e3668..3db24391e31c 100644
@@ -86,6 +86,19 @@ static int spapr_xive_target_to_nvt(uint32_t target,
  * sPAPR END indexing uses a simple mapping of the CPU vcpu_id, 8
  * priorities per CPU
  */
+int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
+                             uint32_t *out_server, uint8_t *out_prio)
+{

Since you don't support irq blocks as yet, should this error out rather than ignoring if end_blk != 0?

yes we could. I will add a test against SPAPR_XIVE_BLOCK which is the value of the sPAPR block ID. I would like to be able to track where it is used even if constant.

Quoted 149 lines
Quoted 149 lines
hw/intc/spapr_xive.c +76 -8
+    if (out_server) {
+        *out_server = end_idx >> 3;
+    }
+
+    if (out_prio) {
+        *out_prio = end_idx & 0x7;
+    }
+    return 0;
+}
+
 static void spapr_xive_cpu_to_end(PowerPCCPU *cpu, uint8_t prio,
                                   uint8_t *out_end_blk, uint32_t *out_end_idx)
 {
@@ -792,6 +805,16 @@ static target_ulong h_int_set_source_config(PowerPCCPU *cpu,
         new_eas.w = xive_set_field64(EAS_END_DATA, new_eas.w, eisn);
     }
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_set_source_config(xive, lisn, &new_eas, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
+
 out:
     xive->eat[lisn] = new_eas;
     return H_SUCCESS;
@@ -1097,6 +1120,16 @@ static target_ulong h_int_set_queue_config(PowerPCCPU *cpu,
      */
 
 out:
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_set_queue_config(xive, end_blk, end_idx, &end, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
+
     /* Update END */
     memcpy(&xive->endt[end_idx], &end, sizeof(XiveEND));
     return H_SUCCESS;
@@ -1189,6 +1222,16 @@ static target_ulong h_int_get_queue_config(PowerPCCPU *cpu,
         args[2] = 0;
     }
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_get_queue_config(xive, end_blk, end_idx, end, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
+
     /* TODO: do we need any locking on the END ? */
     if (flags & SPAPR_XIVE_END_DEBUG) {
         /* Load the event queue generation number into the return flags */
@@ -1341,15 +1384,20 @@ static target_ulong h_int_esb(PowerPCCPU *cpu,
         return H_P3;
     }
 
-    mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
+    if (kvm_irqchip_in_kernel()) {
+        args[0] = kvmppc_xive_esb_rw(xsrc, lisn, offset, data,
+                                     flags & SPAPR_XIVE_ESB_STORE);
+    } else {
+        mmio_addr = xive->vc_base + xive_source_esb_mgmt(xsrc, lisn) + offset;
 
-    if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
-                      (flags & SPAPR_XIVE_ESB_STORE))) {
-        qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
-                      HWADDR_PRIx "\n", mmio_addr);
-        return H_HARDWARE;
+        if (dma_memory_rw(&address_space_memory, mmio_addr, &data, 8,
+                          (flags & SPAPR_XIVE_ESB_STORE))) {
+            qemu_log_mask(LOG_GUEST_ERROR, "XIVE: failed to access ESB @0x%"
+                          HWADDR_PRIx "\n", mmio_addr);
+            return H_HARDWARE;
+        }
+        args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
     }
-    args[0] = (flags & SPAPR_XIVE_ESB_STORE) ? -1 : data;
     return H_SUCCESS;
 }
 
@@ -1406,7 +1454,20 @@ static target_ulong h_int_sync(PowerPCCPU *cpu,
      * This is not needed when running the emulation under QEMU
      */
 
-    /* This is not real hardware. Nothing to be done */
+    /*
+     * This is not real hardware. Nothing to be done unless when
+     * under KVM
+     */
+
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_sync_source(xive, lisn, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
     return H_SUCCESS;
 }
 
@@ -1441,6 +1502,16 @@ static target_ulong h_int_reset(PowerPCCPU *cpu,
     }
 
     device_reset(DEVICE(xive));
+
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_reset(xive, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return H_HARDWARE;
+        }
+    }
     return H_SUCCESS;
 }
 
hw/intc/spapr_xive_kvm.c +9 -0
index 623fbf74f23e..6b50451b4f85 100644
@@ -89,6 +89,52 @@ void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
  * XIVE Interrupt Source (KVM)
  */
 
+void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
+                                   Error **errp)
+{
+    uint32_t end_idx;
+    uint32_t end_blk;
+    uint32_t eisn;
+    uint8_t priority;
+    uint32_t server;
+    uint64_t kvm_src;
Quoted 8 lines

hw/intc/spapr_xive_kvm.c void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp) · line 105

Quoted 5 lines
hw/intc/spapr_xive_kvm.c +5 -0
+    Error *local_err = NULL;
+
+    /*
+     * No need to set a MASKED source, this is the default state after
+     * reset.

I don't quite follow this comment, why is there no need to call a MASKED source?

because MASKED is the default state in which KVM initializes the IRQ. I will clarify.

Quoted 39 lines
Quoted 39 lines
hw/intc/spapr_xive_kvm.c +32 -0
+     */
+    if (!xive_eas_is_valid(eas) || xive_eas_is_masked(eas)) {
+        return;
+    }
+
+    end_idx = xive_get_field64(EAS_END_INDEX, eas->w);
+    end_blk = xive_get_field64(EAS_END_BLOCK, eas->w);
+    eisn = xive_get_field64(EAS_END_DATA, eas->w);
+
+    spapr_xive_end_to_target(end_blk, end_idx, &server, &priority);
+
+    kvm_src = priority << KVM_XIVE_SOURCE_PRIORITY_SHIFT &
+        KVM_XIVE_SOURCE_PRIORITY_MASK;
+    kvm_src |= server << KVM_XIVE_SOURCE_SERVER_SHIFT &
+        KVM_XIVE_SOURCE_SERVER_MASK;
+    kvm_src |= ((uint64_t)eisn << KVM_XIVE_SOURCE_EISN_SHIFT) &
+        KVM_XIVE_SOURCE_EISN_MASK;
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SOURCE_CONFIG, lisn,
+                      &kvm_src, true, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+}
+
+void kvmppc_xive_sync_source(sPAPRXive *xive, uint32_t lisn, Error **errp)
+{
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_SOURCE_SYNC, lisn,
+                      NULL, true, errp);
+}
+
 /*
  * At reset, the interrupt sources are simply created and MASKED. We
  * only need to inform the KVM XIVE device about their type: LSI or
@@ -125,6 +171,64 @@ void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp)
     }
 }
 
Quoted 12 lines

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 182

Quoted 9 lines
hw/intc/spapr_xive_kvm.c +9 -0
+/*
+ * This is used to perform the magic loads on the ESB pages, described
+ * in xive.h.
+ */
+static uint64_t xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write)
+{
+    unsigned long addr = (unsigned long) xsrc->esb_mmap +
+        xive_source_esb_mgmt(xsrc, srcno) + offset;

Casting the esb_mmap into unsigned long then back to a pointer looks unnecessary. You should be able to do this with pointer arithmetic.

yes.

Quoted 9 lines

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 189

Quoted 6 lines
hw/intc/spapr_xive_kvm.c +6 -0
+    if (write) {
+        *((uint64_t *) addr) = data;
+        return -1;
+    } else {
+        return *((uint64_t *) addr);
+    }

Since this is always dealing with 64-bit values, couldn't you put the byteswaps in here rather than in all the callers?

indeed.

Quoted 8 lines

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 195

Quoted 6 lines
hw/intc/spapr_xive_kvm.c +6 -0
+}
+
+static uint8_t xive_esb_read(XiveSource *xsrc, int srcno, uint32_t offset)
+{
+    /* Prevent the compiler from optimizing away the load */
+    volatile uint64_t value = xive_esb_rw(xsrc, srcno, offset, 0, 0);

Wouldn't the volatile magic be better inside xive_esb_rw()?

sure. I will rework these helpers.

Quoted 5 lines
Quoted 5 lines
hw/intc/spapr_xive_kvm.c +5 -0
+    return be64_to_cpu(value) & 0x3;
+}
+
+static void xive_esb_trigger(XiveSource *xsrc, int srcno)
+{
Quoted 8 lines

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp) · line 206

Quoted 5 lines
hw/intc/spapr_xive_kvm.c +5 -0
+    unsigned long addr = (unsigned long) xsrc->esb_mmap +
+        xive_source_esb_page(xsrc, srcno);
+
+    *((uint64_t *) addr) = 0x0;
+}

Also.. aren't some of these register accesses likely to need memory barriers?

AIUI, these are CI pages. So we shouldn't need barriers.

Quoted 37 lines
Quoted 37 lines
hw/intc/spapr_xive_kvm.c +30 -0
+
+uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
+                            uint64_t data, bool write)
+{
+    if (write) {
+        return xive_esb_rw(xsrc, srcno, offset, data, 1);
+    }
+
+    /*
+     * Special Load EOI handling for LSI sources. Q bit is never set
+     * and the interrupt should be re-triggered if the level is still
+     * asserted.
+     */
+    if (xive_source_irq_is_lsi(xsrc, srcno) &&
+        offset == XIVE_ESB_LOAD_EOI) {
+        xive_esb_read(xsrc, srcno, XIVE_ESB_SET_PQ_00);
+        if (xsrc->status[srcno] & XIVE_STATUS_ASSERTED) {
+            xive_esb_trigger(xsrc, srcno);
+        }
+        return 0;
+    } else {
+        return xive_esb_rw(xsrc, srcno, offset, 0, 0);
+    }
+}
+
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 {
     XiveSource *xsrc = opaque;
@@ -155,6 +259,86 @@ void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 /*
  * sPAPR XIVE interrupt controller (KVM)
  */
+void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                  uint32_t end_idx, XiveEND *end,
+                                  Error **errp)
+{
+    struct kvm_ppc_xive_eq kvm_eq = { 0 };
Quoted 8 lines

hw/intc/spapr_xive_kvm.c void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val) · line 272

Quoted 6 lines
hw/intc/spapr_xive_kvm.c +6 -0
+    uint64_t kvm_eq_idx;
+    uint8_t priority;
+    uint32_t server;
+    Error *local_err = NULL;
+
+    if (!xive_end_is_valid(end)) {

This should set an error, shouldn't it?

Hmm, this helper is used in the hcall h_int_get_queue_config() and, later, in kvmppc_xive_get_queues() to synchronize the state from KVM.

I should probably move the test outside this routine, return H_HARDWARE in the hcall and skip invalid ENDs in kvmppc_xive_get_queues()

Thanks,

C.

Quoted 74 lines
Quoted 73 lines
hw/intc/spapr_xive_kvm.c +69 -0
+        return;
+    }
+
+    /* Encode the tuple (server, prio) as a KVM EQ index */
+    spapr_xive_end_to_target(end_blk, end_idx, &server, &priority);
+
+    kvm_eq_idx = priority << KVM_XIVE_EQ_PRIORITY_SHIFT &
+            KVM_XIVE_EQ_PRIORITY_MASK;
+    kvm_eq_idx |= server << KVM_XIVE_EQ_SERVER_SHIFT &
+        KVM_XIVE_EQ_SERVER_MASK;
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_EQ_CONFIG, kvm_eq_idx,
+                      &kvm_eq, false, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
+    /*
+     * The EQ index and toggle bit are updated by HW. These are the
+     * only fields we want to return.
+     */
+    end->w1 = xive_set_field32(END_W1_GENERATION, 0ul, kvm_eq.qtoggle) |
+        xive_set_field32(END_W1_PAGE_OFF, 0ul, kvm_eq.qindex);
+}
+
+void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
+                                  uint32_t end_idx, XiveEND *end,
+                                  Error **errp)
+{
+    struct kvm_ppc_xive_eq kvm_eq = { 0 };
+    uint64_t kvm_eq_idx;
+    uint8_t priority;
+    uint32_t server;
+    Error *local_err = NULL;
+
+    if (!xive_end_is_valid(end)) {
+        return;
+    }
+
+    /* Build the KVM state from the local END structure */
+    kvm_eq.flags   = KVM_XIVE_EQ_FLAG_ALWAYS_NOTIFY;
+    kvm_eq.qsize   = xive_get_field32(END_W0_QSIZE, end->w0) + 12;
+    kvm_eq.qpage   = (uint64_t) be32_to_cpu(end->w2 & 0x0fffffff) << 32 |
+        be32_to_cpu(end->w3);
+    kvm_eq.qtoggle = xive_get_field32(END_W1_GENERATION, end->w1);
+    kvm_eq.qindex  = xive_get_field32(END_W1_PAGE_OFF, end->w1);
+
+    /* Encode the tuple (server, prio) as a KVM EQ index */
+    spapr_xive_end_to_target(end_blk, end_idx, &server, &priority);
+
+    kvm_eq_idx = priority << KVM_XIVE_EQ_PRIORITY_SHIFT &
+            KVM_XIVE_EQ_PRIORITY_MASK;
+    kvm_eq_idx |= server << KVM_XIVE_EQ_SERVER_SHIFT &
+        KVM_XIVE_EQ_SERVER_MASK;
+
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_EQ_CONFIG, kvm_eq_idx,
+                      &kvm_eq, true, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+}
+
+void kvmppc_xive_reset(sPAPRXive *xive, Error **errp)
+{
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_CTRL, KVM_DEV_XIVE_RESET,
+                      NULL, true, errp);
+}
 
 static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
                               Error **errp)
#7 lore raw

PATCH v2 3/13 spapr/xive: activate KVM support

In reply to Cédric Le Goater

All is in place for KVM now. State synchronization and migration will come next.

---
 hw/ppc/spapr_irq.c | 9 ---------
 1 file changed, 9 deletions(-)
hw/ppc/spapr_irq.c +0 -9
index 6e1c36dc62ca..1ad57582a403 100644
@@ -263,19 +263,10 @@ sPAPRIrq spapr_irq_xics = {
 static void spapr_irq_init_xive(sPAPRMachineState *spapr, int nr_irqs,
                                 Error **errp)
 {
-    MachineState *machine = MACHINE(spapr);
     uint32_t nr_servers = spapr_max_server_number(spapr);
     DeviceState *dev;
     int i;
 
-    /* KVM XIVE device not yet available */
-    if (kvm_enabled()) {
-        if (machine_kernel_irqchip_required(machine)) {
-            error_setg(errp, "kernel_irqchip requested. no KVM XIVE support");
-            return;
-        }
-    }
-
     dev = qdev_create(NULL, TYPE_SPAPR_XIVE);
     qdev_prop_set_uint32(dev, "nr-irqs", nr_irqs);
     /*
Signature · 1 line
2.20.1
#8 lore raw

Re: [Qemu-devel] [PATCH v2 03/13] spapr/xive: activate KVM support

Quoted 2 lines · Cédric Le Goater

All is in place for KVM now. State synchronization and migration will come next.

As with the kernel side capability, this should be moved later in the series to avoid breaking bisections.

Quoted 30 lines
---
 hw/ppc/spapr_irq.c | 9 ---------
 1 file changed, 9 deletions(-)
hw/ppc/spapr_irq.c +0 -9
index 6e1c36dc62ca..1ad57582a403 100644
@@ -263,19 +263,10 @@ sPAPRIrq spapr_irq_xics = {
 static void spapr_irq_init_xive(sPAPRMachineState *spapr, int nr_irqs,
                                 Error **errp)
 {
-    MachineState *machine = MACHINE(spapr);
     uint32_t nr_servers = spapr_max_server_number(spapr);
     DeviceState *dev;
     int i;
 
-    /* KVM XIVE device not yet available */
-    if (kvm_enabled()) {
-        if (machine_kernel_irqchip_required(machine)) {
-            error_setg(errp, "kernel_irqchip requested. no KVM XIVE support");
-            return;
-        }
-    }
-
     dev = qdev_create(NULL, TYPE_SPAPR_XIVE);
     qdev_prop_set_uint32(dev, "nr-irqs", nr_irqs);
     /*
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
Reviewed-by
#9 lore raw

Re: [Qemu-devel] [PATCH v2 03/13] spapr/xive: activate KVM support

Quoted 6 lines · David Gibson
Quoted 2 lines · Cédric Le Goater

All is in place for KVM now. State synchronization and migration will come next.

As with the kernel side capability, this should be moved later in the series to avoid breaking bisections.

Apart from that,

Quoted 32 lines
Quoted 31 lines
---
 hw/ppc/spapr_irq.c | 9 ---------
 1 file changed, 9 deletions(-)
hw/ppc/spapr_irq.c +0 -9
index 6e1c36dc62ca..1ad57582a403 100644
@@ -263,19 +263,10 @@ sPAPRIrq spapr_irq_xics = {
 static void spapr_irq_init_xive(sPAPRMachineState *spapr, int nr_irqs,
                                 Error **errp)
 {
-    MachineState *machine = MACHINE(spapr);
     uint32_t nr_servers = spapr_max_server_number(spapr);
     DeviceState *dev;
     int i;
 
-    /* KVM XIVE device not yet available */
-    if (kvm_enabled()) {
-        if (machine_kernel_irqchip_required(machine)) {
-            error_setg(errp, "kernel_irqchip requested. no KVM XIVE support");
-            return;
-        }
-    }
-
     dev = qdev_create(NULL, TYPE_SPAPR_XIVE);
     qdev_prop_set_uint32(dev, "nr-irqs", nr_irqs);
     /*
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
#10 lore raw

Re: [Qemu-devel] [PATCH v2 03/13] spapr/xive: activate KVM support

In reply to David Gibson
Quoted 6 lines · David Gibson
Quoted 2 lines · Cédric Le Goater

All is in place for KVM now. State synchronization and migration will come next.

As with the kernel side capability, this should be moved later in the series to avoid breaking bisections.

I am not sure to understand. At this stage of the patchset, the XIVE exploitation mode is operational. We can not synchronise the state or migrate but it runs.

Should we move XIVE activation after the migration patch ?

C.

Quoted 31 lines
Quoted 30 lines
---
 hw/ppc/spapr_irq.c | 9 ---------
 1 file changed, 9 deletions(-)
hw/ppc/spapr_irq.c +0 -9
index 6e1c36dc62ca..1ad57582a403 100644
@@ -263,19 +263,10 @@ sPAPRIrq spapr_irq_xics = {
 static void spapr_irq_init_xive(sPAPRMachineState *spapr, int nr_irqs,
                                 Error **errp)
 {
-    MachineState *machine = MACHINE(spapr);
     uint32_t nr_servers = spapr_max_server_number(spapr);
     DeviceState *dev;
     int i;
 
-    /* KVM XIVE device not yet available */
-    if (kvm_enabled()) {
-        if (machine_kernel_irqchip_required(machine)) {
-            error_setg(errp, "kernel_irqchip requested. no KVM XIVE support");
-            return;
-        }
-    }
-
     dev = qdev_create(NULL, TYPE_SPAPR_XIVE);
     qdev_prop_set_uint32(dev, "nr-irqs", nr_irqs);
     /*
#11 lore raw

PATCH v2 4/13 spapr/xive: add state synchronization with KVM

In reply to Cédric Le Goater

This extends the KVM XIVE device backend with 'synchronize_state' methods used to retrieve the state from KVM. The HW state of the sources, the KVM device and the thread interrupt contexts are collected for the monitor usage and also migration.

These get operations rely on their KVM counterpart in the host kernel which acts as a proxy for OPAL, the host firmware. The set operations will be added for migration support later.

---
 include/hw/ppc/spapr_xive.h |  8 ++++
 include/hw/ppc/xive.h       |  1 +
 hw/intc/spapr_xive.c        | 17 ++++---
 hw/intc/spapr_xive_kvm.c    | 89 +++++++++++++++++++++++++++++++++++++
 hw/intc/xive.c              | 10 +++++
 5 files changed, 118 insertions(+), 7 deletions(-)
include/hw/ppc/spapr_xive.h +8 -0
index 749c6cbc2c56..ebd65e7fe36b 100644
@@ -44,6 +44,13 @@ typedef struct sPAPRXive {
     void          *tm_mmap;
 } sPAPRXive;
 
+/*
+ * The sPAPR machine has a unique XIVE IC device. Assign a fixed value
+ * to the controller block id value. It can nevertheless be changed
+ * for testing purpose.
+ */
+#define SPAPR_XIVE_BLOCK_ID 0x0
+
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
 bool spapr_xive_irq_free(sPAPRXive *xive, uint32_t lisn);
 void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon);
@@ -74,5 +81,6 @@ void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
 void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
                                  uint32_t end_idx, XiveEND *end,
                                  Error **errp);
+void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +1 -0
index 061d43fea24d..f3766fd881a2 100644
@@ -431,5 +431,6 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp);
 void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
+void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp);
 
 #endif /* PPC_XIVE_H */
hw/intc/spapr_xive.c +10 -7
index 3db24391e31c..9f07567f4d78 100644
@@ -40,13 +40,6 @@
 
 #define SPAPR_XIVE_NVT_BASE 0x400
 
-/*
- * The sPAPR machine has a unique XIVE IC device. Assign a fixed value
- * to the controller block id value. It can nevertheless be changed
- * for testing purpose.
- */
-#define SPAPR_XIVE_BLOCK_ID 0x0
-
 /*
  * sPAPR NVT and END indexing helpers
  */
@@ -153,6 +146,16 @@ void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon)
     XiveSource *xsrc = &xive->source;
     int i;
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_synchronize_state(xive, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return;
+        }
+    }
+
     monitor_printf(mon, "  LSIN         PQ    EISN     CPU/PRIO EQ\n");
 
     for (i = 0; i < xive->nr_irqs; i++) {
hw/intc/spapr_xive_kvm.c +89 -0

107 lines not on this page · raw message

hw/intc/xive.c +10 -0
index 0284b5803551..f478c52ab2a0 100644
@@ -431,6 +431,16 @@ void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
     int i;
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return;
+        }
+    }
+
     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
                    "  W2\n", cpu_index);
 
Signature · 1 line
2.20.1
#12 lore raw

Re: [Qemu-devel] [PATCH v2 04/13] spapr/xive: add state synchronization with KVM

Quoted 108 lines · Cédric Le Goater

This extends the KVM XIVE device backend with 'synchronize_state' methods used to retrieve the state from KVM. The HW state of the sources, the KVM device and the thread interrupt contexts are collected for the monitor usage and also migration.

These get operations rely on their KVM counterpart in the host kernel which acts as a proxy for OPAL, the host firmware. The set operations will be added for migration support later.

---
 include/hw/ppc/spapr_xive.h |  8 ++++
 include/hw/ppc/xive.h       |  1 +
 hw/intc/spapr_xive.c        | 17 ++++---
 hw/intc/spapr_xive_kvm.c    | 89 +++++++++++++++++++++++++++++++++++++
 hw/intc/xive.c              | 10 +++++
 5 files changed, 118 insertions(+), 7 deletions(-)
include/hw/ppc/spapr_xive.h +8 -0
index 749c6cbc2c56..ebd65e7fe36b 100644
@@ -44,6 +44,13 @@ typedef struct sPAPRXive {
     void          *tm_mmap;
 } sPAPRXive;
 
+/*
+ * The sPAPR machine has a unique XIVE IC device. Assign a fixed value
+ * to the controller block id value. It can nevertheless be changed
+ * for testing purpose.
+ */
+#define SPAPR_XIVE_BLOCK_ID 0x0
+
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
 bool spapr_xive_irq_free(sPAPRXive *xive, uint32_t lisn);
 void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon);
@@ -74,5 +81,6 @@ void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
 void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
                                  uint32_t end_idx, XiveEND *end,
                                  Error **errp);
+void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +1 -0
index 061d43fea24d..f3766fd881a2 100644
@@ -431,5 +431,6 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp);
 void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
+void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp);
 
 #endif /* PPC_XIVE_H */
hw/intc/spapr_xive.c +10 -7
index 3db24391e31c..9f07567f4d78 100644
@@ -40,13 +40,6 @@
 
 #define SPAPR_XIVE_NVT_BASE 0x400
 
-/*
- * The sPAPR machine has a unique XIVE IC device. Assign a fixed value
- * to the controller block id value. It can nevertheless be changed
- * for testing purpose.
- */
-#define SPAPR_XIVE_BLOCK_ID 0x0
-
 /*
  * sPAPR NVT and END indexing helpers
  */
@@ -153,6 +146,16 @@ void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon)
     XiveSource *xsrc = &xive->source;
     int i;
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_synchronize_state(xive, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return;
+        }
+    }
+
     monitor_printf(mon, "  LSIN         PQ    EISN     CPU/PRIO EQ\n");
 
     for (i = 0; i < xive->nr_irqs; i++) {
hw/intc/spapr_xive_kvm.c +11 -0
index 6b50451b4f85..4b1ffb9835f9 100644
@@ -60,6 +60,57 @@ static void kvm_cpu_enable(CPUState *cs)
 /*
  * XIVE Thread Interrupt Management context (KVM)
  */
+static void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp)
+{
+    uint64_t state[4] = { 0 };
+    int ret;
+
+    ret = kvm_get_one_reg(tctx->cs, KVM_REG_PPC_NVT_STATE, state);
+    if (ret != 0) {
+        error_setg_errno(errp, errno,
+                         "XIVE: could not capture KVM state of CPU %ld",
+                         kvm_arch_vcpu_id(tctx->cs));
+        return;

hw/intc/spapr_xive_kvm.c static void kvm_cpu_enable(CPUState *cs) · line 83

Quoted 10 lines
hw/intc/spapr_xive_kvm.c +10 -0
+    }
+
+    /* word0 and word1 of the OS ring. */
+    *((uint64_t *) &tctx->regs[TM_QW1_OS]) = state[0];
+
+    /*
+     * KVM also returns word2 containing the OS CAM line which is
+     * interesting to print out in the QEMU monitor.
+     */
+    *((uint64_t *) &tctx->regs[TM_QW1_OS + TM_WORD2]) = state[1];

As mentioned elsewhere, it is interesting for debugging, but doesn't seem to really match the guest visible CAM state, so I'm not convinced it's a good idea to put it into the regs[] structure.

Quoted 14 lines
hw/intc/spapr_xive_kvm.c +14 -0
+}
+
+typedef struct {
+    XiveTCTX *tctx;
+    Error *err;
+} XiveCpuGetState;
+
+static void kvmppc_xive_cpu_do_synchronize_state(CPUState *cpu,
+                                                 run_on_cpu_data arg)
+{
+    XiveCpuGetState *s = arg.host_ptr;
+
+    kvmppc_xive_cpu_get_state(s->tctx, &s->err);
+}

hw/intc/spapr_xive_kvm.c static void kvm_cpu_enable(CPUState *cs) · line 107

Quoted 10 lines
hw/intc/spapr_xive_kvm.c +10 -0
+
+void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp)
+{
+    XiveCpuGetState s = {
+        .tctx = tctx,
+        .err = NULL,
+    };
+
+    run_on_cpu(tctx->cs, kvmppc_xive_cpu_do_synchronize_state,
+               RUN_ON_CPU_HOST_PTR(&s));

Why does this need a run_on_cpu() ? The KVM call which is getting the actual info takes a cpu parameter.

Quoted 82 lines
hw/intc/spapr_xive_kvm.c +44 -0
+
+    if (s.err) {
+        error_propagate(errp, s.err);
+        return;
+    }
+}
 
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
 {
@@ -229,6 +280,19 @@ uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
     }
 }
 
+static void kvmppc_xive_source_get_state(XiveSource *xsrc)
+{
+    int i;
+
+    for (i = 0; i < xsrc->nr_irqs; i++) {
+        /* Perform a load without side effect to retrieve the PQ bits */
+        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+
+        /* and save PQ locally */
+        xive_source_esb_set(xsrc, i, pq);
+    }
+}
+
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 {
     XiveSource *xsrc = opaque;
@@ -340,6 +404,31 @@ void kvmppc_xive_reset(sPAPRXive *xive, Error **errp)
                       NULL, true, errp);
 }
 
+static void kvmppc_xive_get_queues(sPAPRXive *xive, Error **errp)
+{
+    Error *local_err = NULL;
+    int i;
+
+    for (i = 0; i < xive->nr_ends; i++) {
+        kvmppc_xive_get_queue_config(xive, SPAPR_XIVE_BLOCK_ID, i,
+                                     &xive->endt[i], &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+}
+
+void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
+{
+    kvmppc_xive_source_get_state(&xive->source);
+
+    /* EAT: there is no extra state to query from KVM */
+
+    /* ENDT */
+    kvmppc_xive_get_queues(xive, errp);
+}
+
 static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
                               Error **errp)
 {
hw/intc/xive.c +10 -0
index 0284b5803551..f478c52ab2a0 100644
@@ -431,6 +431,16 @@ void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
     int i;
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return;
+        }
+    }
+
     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
                    "  W2\n", cpu_index);
 
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
#13 lore raw

Re: [Qemu-devel] [PATCH v2 04/13] spapr/xive: add state synchronization with KVM

Quoted 114 lines · David Gibson
Quoted 113 lines · Cédric Le Goater

This extends the KVM XIVE device backend with 'synchronize_state' methods used to retrieve the state from KVM. The HW state of the sources, the KVM device and the thread interrupt contexts are collected for the monitor usage and also migration.

These get operations rely on their KVM counterpart in the host kernel which acts as a proxy for OPAL, the host firmware. The set operations will be added for migration support later.

---
 include/hw/ppc/spapr_xive.h |  8 ++++
 include/hw/ppc/xive.h       |  1 +
 hw/intc/spapr_xive.c        | 17 ++++---
 hw/intc/spapr_xive_kvm.c    | 89 +++++++++++++++++++++++++++++++++++++
 hw/intc/xive.c              | 10 +++++
 5 files changed, 118 insertions(+), 7 deletions(-)
include/hw/ppc/spapr_xive.h +8 -0
index 749c6cbc2c56..ebd65e7fe36b 100644
@@ -44,6 +44,13 @@ typedef struct sPAPRXive {
     void          *tm_mmap;
 } sPAPRXive;
 
+/*
+ * The sPAPR machine has a unique XIVE IC device. Assign a fixed value
+ * to the controller block id value. It can nevertheless be changed
+ * for testing purpose.
+ */
+#define SPAPR_XIVE_BLOCK_ID 0x0
+
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
 bool spapr_xive_irq_free(sPAPRXive *xive, uint32_t lisn);
 void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon);
@@ -74,5 +81,6 @@ void kvmppc_xive_set_queue_config(sPAPRXive *xive, uint8_t end_blk,
 void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
                                  uint32_t end_idx, XiveEND *end,
                                  Error **errp);
+void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +1 -0
index 061d43fea24d..f3766fd881a2 100644
@@ -431,5 +431,6 @@ void kvmppc_xive_source_reset_one(XiveSource *xsrc, int srcno, Error **errp);
 void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
+void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp);
 
 #endif /* PPC_XIVE_H */
hw/intc/spapr_xive.c +10 -7
index 3db24391e31c..9f07567f4d78 100644
@@ -40,13 +40,6 @@
 
 #define SPAPR_XIVE_NVT_BASE 0x400
 
-/*
- * The sPAPR machine has a unique XIVE IC device. Assign a fixed value
- * to the controller block id value. It can nevertheless be changed
- * for testing purpose.
- */
-#define SPAPR_XIVE_BLOCK_ID 0x0
-
 /*
  * sPAPR NVT and END indexing helpers
  */
@@ -153,6 +146,16 @@ void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon)
     XiveSource *xsrc = &xive->source;
     int i;
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_synchronize_state(xive, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return;
+        }
+    }
+
     monitor_printf(mon, "  LSIN         PQ    EISN     CPU/PRIO EQ\n");
 
     for (i = 0; i < xive->nr_irqs; i++) {
hw/intc/spapr_xive_kvm.c +16 -0
index 6b50451b4f85..4b1ffb9835f9 100644
@@ -60,6 +60,57 @@ static void kvm_cpu_enable(CPUState *cs)
 /*
  * XIVE Thread Interrupt Management context (KVM)
  */
+static void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp)
+{
+    uint64_t state[4] = { 0 };
+    int ret;
+
+    ret = kvm_get_one_reg(tctx->cs, KVM_REG_PPC_NVT_STATE, state);
+    if (ret != 0) {
+        error_setg_errno(errp, errno,
+                         "XIVE: could not capture KVM state of CPU %ld",
+                         kvm_arch_vcpu_id(tctx->cs));
+        return;
+    }
+
+    /* word0 and word1 of the OS ring. */
+    *((uint64_t *) &tctx->regs[TM_QW1_OS]) = state[0];
+
Quoted 8 lines

hw/intc/spapr_xive_kvm.c static void kvm_cpu_enable(CPUState *cs) · line 83

Quoted 5 lines
hw/intc/spapr_xive_kvm.c +5 -0
+    /*
+     * KVM also returns word2 containing the OS CAM line which is
+     * interesting to print out in the QEMU monitor.
+     */
+    *((uint64_t *) &tctx->regs[TM_QW1_OS + TM_WORD2]) = state[1];

As mentioned elsewhere, it is interesting for debugging, but doesn't seem to really match the guest visible CAM state,

The guest is not allowed to see these registers in the TIMA OS page and we are not violating the XIVE architecture. That is where the CAM value belong in HW. The exact same place. I was even thinking to propagate the POOL value which could be useful for nested.

Quoted 2 lines

so I'm not convinced it's a good idea to put it into the regs[] structure.

I understand it is problematic in case of a KVM->QEMU migration because we need to force a XiveTCTX reset to update the registers with the QEMU CAM line which has been overridden with the KVM CAM line.

Another solution could be to add a 'nvt_base' property to SpaprXive and a KVM control to get its value (xive->vp_base in the KVM XIVE device). It would be migrated and used by the QEMU XIVE device after migration.

Quoted 19 lines
Quoted 19 lines
hw/intc/spapr_xive_kvm.c +19 -0
+}
+
+typedef struct {
+    XiveTCTX *tctx;
+    Error *err;
+} XiveCpuGetState;
+
+static void kvmppc_xive_cpu_do_synchronize_state(CPUState *cpu,
+                                                 run_on_cpu_data arg)
+{
+    XiveCpuGetState *s = arg.host_ptr;
+
+    kvmppc_xive_cpu_get_state(s->tctx, &s->err);
+}
+
+void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp)
+{
+    XiveCpuGetState s = {
+        .tctx = tctx,
Quoted 8 lines

hw/intc/spapr_xive_kvm.c static void kvm_cpu_enable(CPUState *cs) · line 107

Quoted 5 lines
hw/intc/spapr_xive_kvm.c +5 -0
+        .err = NULL,
+    };
+
+    run_on_cpu(tctx->cs, kvmppc_xive_cpu_do_synchronize_state,
+               RUN_ON_CPU_HOST_PTR(&s));

Why does this need a run_on_cpu() ? The KVM call which is getting the actual info takes a cpu parameter.

Don't we need to kick the vCPU ?

Thanks,

C.

Quoted 84 lines
Quoted 83 lines
hw/intc/spapr_xive_kvm.c +44 -0
+
+    if (s.err) {
+        error_propagate(errp, s.err);
+        return;
+    }
+}
 
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
 {
@@ -229,6 +280,19 @@ uint64_t kvmppc_xive_esb_rw(XiveSource *xsrc, int srcno, uint32_t offset,
     }
 }
 
+static void kvmppc_xive_source_get_state(XiveSource *xsrc)
+{
+    int i;
+
+    for (i = 0; i < xsrc->nr_irqs; i++) {
+        /* Perform a load without side effect to retrieve the PQ bits */
+        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+
+        /* and save PQ locally */
+        xive_source_esb_set(xsrc, i, pq);
+    }
+}
+
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 {
     XiveSource *xsrc = opaque;
@@ -340,6 +404,31 @@ void kvmppc_xive_reset(sPAPRXive *xive, Error **errp)
                       NULL, true, errp);
 }
 
+static void kvmppc_xive_get_queues(sPAPRXive *xive, Error **errp)
+{
+    Error *local_err = NULL;
+    int i;
+
+    for (i = 0; i < xive->nr_ends; i++) {
+        kvmppc_xive_get_queue_config(xive, SPAPR_XIVE_BLOCK_ID, i,
+                                     &xive->endt[i], &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+}
+
+void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
+{
+    kvmppc_xive_source_get_state(&xive->source);
+
+    /* EAT: there is no extra state to query from KVM */
+
+    /* ENDT */
+    kvmppc_xive_get_queues(xive, errp);
+}
+
 static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
                               Error **errp)
 {
hw/intc/xive.c +10 -0
index 0284b5803551..f478c52ab2a0 100644
@@ -431,6 +431,16 @@ void xive_tctx_pic_print_info(XiveTCTX *tctx, Monitor *mon)
     int cpu_index = tctx->cs ? tctx->cs->cpu_index : -1;
     int i;
 
+    if (kvm_irqchip_in_kernel()) {
+        Error *local_err = NULL;
+
+        kvmppc_xive_cpu_synchronize_state(tctx, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return;
+        }
+    }
+
     monitor_printf(mon, "CPU[%04x]:   QW   NSR CPPR IPB LSMFB ACK# INC AGE PIPR"
                    "  W2\n", cpu_index);
 
#14 lore raw

PATCH v2 5/13 spapr/xive: introduce a VM state change handler

In reply to Cédric Le Goater

This handler is in charge of stabilizing the flow of event notifications in the XIVE controller before migrating a guest. This is a requirement before transferring the guest EQ pages to a destination.

When the VM is stopped, the handler masks the sources (PQ=01) to stop the flow of events and saves their previous state. The XIVE controller is then synced through KVM to flush any in-flight event notification and to stabilize the EQs. At this stage, the EQ pages are marked dirty to make sure the EQ pages are transferred if a migration sequence is in progress.

The previous configuration of the sources is restored when the VM resumes, after a migration or a stop.

---
 include/hw/ppc/spapr_xive.h |  1 +
 hw/intc/spapr_xive_kvm.c    | 77 ++++++++++++++++++++++++++++++++++++-
 2 files changed, 77 insertions(+), 1 deletion(-)
include/hw/ppc/spapr_xive.h +1 -0
index ebd65e7fe36b..298d204d54ef 100644
@@ -42,6 +42,7 @@ typedef struct sPAPRXive {
     /* KVM support */
     int           fd;
     void          *tm_mmap;
+    VMChangeStateEntry *change;
 } sPAPRXive;
 
 /*
hw/intc/spapr_xive_kvm.c +76 -1

91 lines not on this page · raw message

Signature · 1 line
2.20.1
Reviewed-by on v2 5/13
#15 lore raw

Re: [Qemu-devel] [PATCH v2 05/13] spapr/xive: introduce a VM state change handler

Quoted 11 lines · Cédric Le Goater

This handler is in charge of stabilizing the flow of event notifications in the XIVE controller before migrating a guest. This is a requirement before transferring the guest EQ pages to a destination.

When the VM is stopped, the handler masks the sources (PQ=01) to stop the flow of events and saves their previous state. The XIVE controller is then synced through KVM to flush any in-flight event notification and to stabilize the EQs. At this stage, the EQ pages are marked dirty to make sure the EQ pages are transferred if a migration sequence is in progress.

Quoted 4 lines

The previous configuration of the sources is restored when the VM resumes, after a migration or a stop.

Quoted 114 lines
---
 include/hw/ppc/spapr_xive.h |  1 +
 hw/intc/spapr_xive_kvm.c    | 77 ++++++++++++++++++++++++++++++++++++-
 2 files changed, 77 insertions(+), 1 deletion(-)
include/hw/ppc/spapr_xive.h +1 -0
index ebd65e7fe36b..298d204d54ef 100644
@@ -42,6 +42,7 @@ typedef struct sPAPRXive {
     /* KVM support */
     int           fd;
     void          *tm_mmap;
+    VMChangeStateEntry *change;
 } sPAPRXive;
 
 /*
hw/intc/spapr_xive_kvm.c +76 -1
index 4b1ffb9835f9..44d80175b1b5 100644
@@ -419,9 +419,81 @@ static void kvmppc_xive_get_queues(sPAPRXive *xive, Error **errp)
     }
 }
 
+/*
+ * The primary goal of the XIVE VM change handler is to mark the EQ
+ * pages dirty when all XIVE event notifications have stopped.
+ *
+ * Whenever the VM is stopped, the VM change handler masks the sources
+ * (PQ=01) to stop the flow of events and saves the previous state in
+ * anticipation of a migration. The XIVE controller is then synced
+ * through KVM to flush any in-flight event notification and stabilize
+ * the EQs.
+ *
+ * At this stage, we can mark the EQ page dirty and let a migration
+ * sequence transfer the EQ pages to the destination, which is done
+ * just after the stop state.
+ *
+ * The previous configuration of the sources is restored when the VM
+ * runs again.
+ */
+static void kvmppc_xive_change_state_handler(void *opaque, int running,
+                                             RunState state)
+{
+    sPAPRXive *xive = opaque;
+    XiveSource *xsrc = &xive->source;
+    Error *local_err = NULL;
+    int i;
+
+    /*
+     * Restore the sources to their initial state. This is called when
+     * the VM resumes after a stop or a migration.
+     */
+    if (running) {
+        for (i = 0; i < xsrc->nr_irqs; i++) {
+            uint8_t pq = xive_source_esb_get(xsrc, i);
+            if (xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8)) != 0x1) {
+                error_report("XIVE: IRQ %d has an invalid state", i);
+            }
+        }
+
+        return;
+    }
+
+    /*
+     * Mask the sources, to stop the flow of event notifications, and
+     * save the PQs locally in the XiveSource object. The XiveSource
+     * state will be collected later on by its vmstate handler if a
+     * migration is in progress.
+     */
+    for (i = 0; i < xsrc->nr_irqs; i++) {
+        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_01);
+        xive_source_esb_set(xsrc, i, pq);
+    }
+
+    /*
+     * Sync the XIVE controller in KVM, to flush in-flight event
+     * notification that should be enqueued in the EQs and mark the
+     * XIVE EQ pages dirty to collect all updates.
+     */
+    kvm_device_access(xive->fd, KVM_DEV_XIVE_GRP_CTRL,
+                      KVM_DEV_XIVE_EQ_SYNC, NULL, true, &local_err);
+    if (local_err) {
+        error_report_err(local_err);
+        return;
+    }
+}
+
 void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
 {
-    kvmppc_xive_source_get_state(&xive->source);
+    /*
+     * When the VM is stopped, the sources are masked and the previous
+     * state is saved in anticipation of a migration. We should not
+     * synchronize the source state in that case else we will override
+     * the saved state.
+     */
+    if (runstate_is_running()) {
+        kvmppc_xive_source_get_state(&xive->source);
+    }
 
     /* EAT: there is no extra state to query from KVM */
 
@@ -501,6 +573,9 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
                                       "xive.tima", tima_len, xive->tm_mmap);
     sysbus_init_mmio(SYS_BUS_DEVICE(xive), &xive->tm_mmio);
 
+    xive->change = qemu_add_vm_change_state_handler(
+        kvmppc_xive_change_state_handler, xive);
+
     kvm_kernel_irqchip = true;
     kvm_msi_via_irqfd_allowed = true;
     kvm_gsi_direct_mapping = true;
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
#16 lore raw

PATCH v2 6/13 spapr/xive: add migration support for KVM

In reply to Cédric Le Goater

When the VM is stopped, the VM state handler stabilizes the XIVE IC and marks the EQ pages dirty. These are then transferred to destination before the transfer of the device vmstates starts.

The sPAPRXive interrupt controller model captures the XIVE internal tables, EAT and ENDT and the XiveTCTX model does the same for the thread interrupt context registers.

At restart, the sPAPRXive 'post_load' method restores all the XIVE states. It is called by the sPAPR machine 'post_load' method, when all XIVE states have been transferred and loaded.

Finally, the source states are restored in the VM change state handler when the machine reaches the running state.

---
 include/hw/ppc/spapr_xive.h |  3 ++
 include/hw/ppc/xive.h       |  1 +
 hw/intc/spapr_xive.c        | 24 ++++++++++
 hw/intc/spapr_xive_kvm.c    | 93 ++++++++++++++++++++++++++++++++++++-
 hw/intc/xive.c              | 17 +++++++
 hw/ppc/spapr_irq.c          |  2 +-
 6 files changed, 138 insertions(+), 2 deletions(-)
include/hw/ppc/spapr_xive.h +3 -0
index 298d204d54ef..22d70650b51f 100644
@@ -55,6 +55,7 @@ typedef struct sPAPRXive {
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
 bool spapr_xive_irq_free(sPAPRXive *xive, uint32_t lisn);
 void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon);
+int spapr_xive_post_load(sPAPRXive *xive, int version_id);
 
 void spapr_xive_hcall_init(sPAPRMachineState *spapr);
 void spapr_dt_xive(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
@@ -83,5 +84,7 @@ void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
                                  uint32_t end_idx, XiveEND *end,
                                  Error **errp);
 void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp);
+int kvmppc_xive_pre_save(sPAPRXive *xive);
+int kvmppc_xive_post_load(sPAPRXive *xive, int version_id);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +1 -0
index f3766fd881a2..3b1baa783975 100644
@@ -432,5 +432,6 @@ void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
 void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp);
+void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp);
 
 #endif /* PPC_XIVE_H */
hw/intc/spapr_xive.c +24 -0
index 9f07567f4d78..21fe5e1aa39f 100644
@@ -469,10 +469,34 @@ static const VMStateDescription vmstate_spapr_xive_eas = {
     },
 };
 
+static int vmstate_spapr_xive_pre_save(void *opaque)
+{
+    if (kvm_irqchip_in_kernel()) {
+        return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
+    }
+
+    return 0;
+}
+
+/*
+ * Called by the sPAPR IRQ backend 'post_load' method at the machine
+ * level.
+ */
+int spapr_xive_post_load(sPAPRXive *xive, int version_id)
+{
+    if (kvm_irqchip_in_kernel()) {
+        return kvmppc_xive_post_load(xive, version_id);
+    }
+
+    return 0;
+}
+
 static const VMStateDescription vmstate_spapr_xive = {
     .name = TYPE_SPAPR_XIVE,
     .version_id = 1,
     .minimum_version_id = 1,
+    .pre_save = vmstate_spapr_xive_pre_save,
+    .post_load = NULL, /* handled at the machine level */
     .fields = (VMStateField[]) {
         VMSTATE_UINT32_EQUAL(nr_irqs, sPAPRXive, NULL),
         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, sPAPRXive, nr_irqs,
hw/intc/spapr_xive_kvm.c +92 -1

111 lines not on this page · raw message

hw/intc/xive.c +17 -0
index f478c52ab2a0..1f8e923ca654 100644
@@ -518,10 +518,27 @@ static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
     qemu_unregister_reset(xive_tctx_reset, dev);
 }
 
+static int vmstate_xive_tctx_pre_save(void *opaque)
+{
+    Error *local_err = NULL;
+
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return -1;
+        }
+    }
+
+    return 0;
+}
+
 static const VMStateDescription vmstate_xive_tctx = {
     .name = TYPE_XIVE_TCTX,
     .version_id = 1,
     .minimum_version_id = 1,
+    .pre_save = vmstate_xive_tctx_pre_save,
+    .post_load = NULL, /* handled by the sPAPRxive model */
     .fields = (VMStateField[]) {
         VMSTATE_BUFFER(regs, XiveTCTX),
         VMSTATE_END_OF_LIST()
hw/ppc/spapr_irq.c +1 -1
index 1ad57582a403..12ecca6264f3 100644
@@ -356,7 +356,7 @@ static void spapr_irq_cpu_intc_create_xive(sPAPRMachineState *spapr,
 
 static int spapr_irq_post_load_xive(sPAPRMachineState *spapr, int version_id)
 {
-    return 0;
+    return spapr_xive_post_load(spapr->xive, version_id);
 }
 
 static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
Signature · 1 line
2.20.1
Reviewed-by on v2 6/13
#17 lore raw

Re: [Qemu-devel] [PATCH v2 06/13] spapr/xive: add migration support for KVM

Quoted 8 lines · Cédric Le Goater

When the VM is stopped, the VM state handler stabilizes the XIVE IC and marks the EQ pages dirty. These are then transferred to destination before the transfer of the device vmstates starts.

The sPAPRXive interrupt controller model captures the XIVE internal tables, EAT and ENDT and the XiveTCTX model does the same for the thread interrupt context registers.

Quoted 8 lines

At restart, the sPAPRXive 'post_load' method restores all the XIVE states. It is called by the sPAPR machine 'post_load' method, when all XIVE states have been transferred and loaded.

Finally, the source states are restored in the VM change state handler when the machine reaches the running state.

Quoted 242 lines
---
 include/hw/ppc/spapr_xive.h |  3 ++
 include/hw/ppc/xive.h       |  1 +
 hw/intc/spapr_xive.c        | 24 ++++++++++
 hw/intc/spapr_xive_kvm.c    | 93 ++++++++++++++++++++++++++++++++++++-
 hw/intc/xive.c              | 17 +++++++
 hw/ppc/spapr_irq.c          |  2 +-
 6 files changed, 138 insertions(+), 2 deletions(-)
include/hw/ppc/spapr_xive.h +3 -0
index 298d204d54ef..22d70650b51f 100644
@@ -55,6 +55,7 @@ typedef struct sPAPRXive {
 bool spapr_xive_irq_claim(sPAPRXive *xive, uint32_t lisn, bool lsi);
 bool spapr_xive_irq_free(sPAPRXive *xive, uint32_t lisn);
 void spapr_xive_pic_print_info(sPAPRXive *xive, Monitor *mon);
+int spapr_xive_post_load(sPAPRXive *xive, int version_id);
 
 void spapr_xive_hcall_init(sPAPRMachineState *spapr);
 void spapr_dt_xive(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
@@ -83,5 +84,7 @@ void kvmppc_xive_get_queue_config(sPAPRXive *xive, uint8_t end_blk,
                                  uint32_t end_idx, XiveEND *end,
                                  Error **errp);
 void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp);
+int kvmppc_xive_pre_save(sPAPRXive *xive);
+int kvmppc_xive_post_load(sPAPRXive *xive, int version_id);
 
 #endif /* PPC_SPAPR_XIVE_H */
include/hw/ppc/xive.h +1 -0
index f3766fd881a2..3b1baa783975 100644
@@ -432,5 +432,6 @@ void kvmppc_xive_source_reset(XiveSource *xsrc, Error **errp);
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val);
 void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp);
 void kvmppc_xive_cpu_synchronize_state(XiveTCTX *tctx, Error **errp);
+void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp);
 
 #endif /* PPC_XIVE_H */
hw/intc/spapr_xive.c +24 -0
index 9f07567f4d78..21fe5e1aa39f 100644
@@ -469,10 +469,34 @@ static const VMStateDescription vmstate_spapr_xive_eas = {
     },
 };
 
+static int vmstate_spapr_xive_pre_save(void *opaque)
+{
+    if (kvm_irqchip_in_kernel()) {
+        return kvmppc_xive_pre_save(SPAPR_XIVE(opaque));
+    }
+
+    return 0;
+}
+
+/*
+ * Called by the sPAPR IRQ backend 'post_load' method at the machine
+ * level.
+ */
+int spapr_xive_post_load(sPAPRXive *xive, int version_id)
+{
+    if (kvm_irqchip_in_kernel()) {
+        return kvmppc_xive_post_load(xive, version_id);
+    }
+
+    return 0;
+}
+
 static const VMStateDescription vmstate_spapr_xive = {
     .name = TYPE_SPAPR_XIVE,
     .version_id = 1,
     .minimum_version_id = 1,
+    .pre_save = vmstate_spapr_xive_pre_save,
+    .post_load = NULL, /* handled at the machine level */
     .fields = (VMStateField[]) {
         VMSTATE_UINT32_EQUAL(nr_irqs, sPAPRXive, NULL),
         VMSTATE_STRUCT_VARRAY_POINTER_UINT32(eat, sPAPRXive, nr_irqs,
hw/intc/spapr_xive_kvm.c +92 -1
index 44d80175b1b5..119fd59fc9ae 100644
@@ -15,6 +15,7 @@
 #include "sysemu/cpus.h"
 #include "sysemu/kvm.h"
 #include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_cpu_core.h"
 #include "hw/ppc/spapr_xive.h"
 #include "hw/ppc/xive.h"
 #include "kvm_ppc.h"
@@ -60,7 +61,30 @@ static void kvm_cpu_enable(CPUState *cs)
 /*
  * XIVE Thread Interrupt Management context (KVM)
  */
-static void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp)
+
+static void kvmppc_xive_cpu_set_state(XiveTCTX *tctx, Error **errp)
+{
+    uint64_t state[4];
+    int ret;
+
+    /* word0 and word1 of the OS ring. */
+    state[0] = *((uint64_t *) &tctx->regs[TM_QW1_OS]);
+
+    /*
+     * OS CAM line. Used by KVM to print out the VP identifier. This
+     * is for debug only.
+     */
+    state[1] = *((uint64_t *) &tctx->regs[TM_QW1_OS + TM_WORD2]);
+
+    ret = kvm_set_one_reg(tctx->cs, KVM_REG_PPC_NVT_STATE, state);
+    if (ret != 0) {
+        error_setg_errno(errp, errno,
+                         "XIVE: could not restore KVM state of CPU %ld",
+                         kvm_arch_vcpu_id(tctx->cs));
+    }
+}
+
+void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp)
 {
     uint64_t state[4] = { 0 };
     int ret;
@@ -501,6 +525,73 @@ void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
     kvmppc_xive_get_queues(xive, errp);
 }
 
+/*
+ * The sPAPRXive 'pre_save' method is called by the vmstate handler of
+ * the sPAPRXive model, after the XIVE controller is synced in the VM
+ * change handler.
+ */
+int kvmppc_xive_pre_save(sPAPRXive *xive)
+{
+    Error *local_err = NULL;
+
+    /* EAT: there is no extra state to query from KVM */
+
+    /* ENDT */
+    kvmppc_xive_get_queues(xive, &local_err);
+    if (local_err) {
+        error_report_err(local_err);
+        return -1;
+    }
+
+    return 0;
+}
+
+/*
+ * The sPAPRXive 'post_load' method is not called by a vmstate
+ * handler. It is called at the sPAPR machine level at the end of the
+ * migration sequence by the sPAPR IRQ backend 'post_load' method,
+ * when all XIVE states have been transferred and loaded.
+ */
+int kvmppc_xive_post_load(sPAPRXive *xive, int version_id)
+{
+    Error *local_err = NULL;
+    CPUState *cs;
+    int i;
+
+    /* Restore the ENDT first. The targetting depends on it. */
+    for (i = 0; i < xive->nr_ends; i++) {
+        kvmppc_xive_set_queue_config(xive, SPAPR_XIVE_BLOCK_ID, i,
+                                     &xive->endt[i], &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return -1;
+        }
+    }
+
+    /* Restore the EAT */
+    for (i = 0; i < xive->nr_irqs; i++) {
+        kvmppc_xive_set_source_config(xive, i, &xive->eat[i], &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return -1;
+        }
+    }
+
+    /* Restore the thread interrupt contexts */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        kvmppc_xive_cpu_set_state(spapr_cpu_state(cpu)->tctx, &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return -1;
+        }
+    }
+
+    /* The source states will be restored when the machine starts running */
+    return 0;
+}
+
 static void *kvmppc_xive_mmap(sPAPRXive *xive, int pgoff, size_t len,
                               Error **errp)
 {
hw/intc/xive.c +17 -0
index f478c52ab2a0..1f8e923ca654 100644
@@ -518,10 +518,27 @@ static void xive_tctx_unrealize(DeviceState *dev, Error **errp)
     qemu_unregister_reset(xive_tctx_reset, dev);
 }
 
+static int vmstate_xive_tctx_pre_save(void *opaque)
+{
+    Error *local_err = NULL;
+
+    if (kvm_irqchip_in_kernel()) {
+        kvmppc_xive_cpu_get_state(XIVE_TCTX(opaque), &local_err);
+        if (local_err) {
+            error_report_err(local_err);
+            return -1;
+        }
+    }
+
+    return 0;
+}
+
 static const VMStateDescription vmstate_xive_tctx = {
     .name = TYPE_XIVE_TCTX,
     .version_id = 1,
     .minimum_version_id = 1,
+    .pre_save = vmstate_xive_tctx_pre_save,
+    .post_load = NULL, /* handled by the sPAPRxive model */
     .fields = (VMStateField[]) {
         VMSTATE_BUFFER(regs, XiveTCTX),
         VMSTATE_END_OF_LIST()
hw/ppc/spapr_irq.c +1 -1
index 1ad57582a403..12ecca6264f3 100644
@@ -356,7 +356,7 @@ static void spapr_irq_cpu_intc_create_xive(sPAPRMachineState *spapr,
 
 static int spapr_irq_post_load_xive(sPAPRMachineState *spapr, int version_id)
 {
-    return 0;
+    return spapr_xive_post_load(spapr->xive, version_id);
 }
 
 static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
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
#18 lore raw

PATCH v2 7/13 spapr/xive: fix migration of the XiveTCTX under TCG

In reply to Cédric Le Goater

When the thread interrupt management state is retrieved from the KVM VCPU, word2 is saved under the QEMU XIVE thread context to print out the OS CAM line under the QEMU monitor.

This breaks the migration of a TCG guest (and with KVM when kernel_irqchip=off) because the matching algorithm of the presenter relies on the OS CAM value. Fix with an extra reset of the thread contexts to restore the expected value.

---
 hw/ppc/spapr_irq.c | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)
hw/ppc/spapr_irq.c +25 -1
index 12ecca6264f3..3176098b9f7c 100644
@@ -356,7 +356,31 @@ static void spapr_irq_cpu_intc_create_xive(sPAPRMachineState *spapr,
 
 static int spapr_irq_post_load_xive(sPAPRMachineState *spapr, int version_id)
 {
-    return spapr_xive_post_load(spapr->xive, version_id);
+    CPUState *cs;
+    int ret;
+
+    ret = spapr_xive_post_load(spapr->xive, version_id);
+    if (ret) {
+        return ret;
+    }
+
+    /*
+     * When the states are collected from the KVM XIVE device, word2
+     * of the XiveTCTX is set to print out the OS CAM line under the
+     * QEMU monitor.
+     *
+     * This breaks the migration on a TCG guest (or on KVM with
+     * kernel_irqchip=off) because the matching algorithm of the
+     * presenter relies on the OS CAM value. Fix with an extra reset
+     * of the thread contexts to restore the expected value.
+     */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        /* (TCG) Set the OS CAM line of the thread interrupt context. */
+        spapr_xive_set_tctx_os_cam(spapr_cpu_state(cpu)->tctx);
+    }
+    return 0;
 }
 
 static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
Signature · 1 line
2.20.1
#19 lore raw

Re: [Qemu-devel] [PATCH v2 07/13] spapr/xive: fix migration of the XiveTCTX under TCG

Quoted 10 lines · Cédric Le Goater

When the thread interrupt management state is retrieved from the KVM VCPU, word2 is saved under the QEMU XIVE thread context to print out the OS CAM line under the QEMU monitor.

This breaks the migration of a TCG guest (and with KVM when kernel_irqchip=off) because the matching algorithm of the presenter relies on the OS CAM value. Fix with an extra reset of the thread contexts to restore the expected value.

As noted elsewhere, I'm not sure this is the right approach to fixing this. In any case this can be folded into the previous patch.

Quoted 41 lines
---
 hw/ppc/spapr_irq.c | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)
hw/ppc/spapr_irq.c +25 -1
index 12ecca6264f3..3176098b9f7c 100644
@@ -356,7 +356,31 @@ static void spapr_irq_cpu_intc_create_xive(sPAPRMachineState *spapr,
 
 static int spapr_irq_post_load_xive(sPAPRMachineState *spapr, int version_id)
 {
-    return spapr_xive_post_load(spapr->xive, version_id);
+    CPUState *cs;
+    int ret;
+
+    ret = spapr_xive_post_load(spapr->xive, version_id);
+    if (ret) {
+        return ret;
+    }
+
+    /*
+     * When the states are collected from the KVM XIVE device, word2
+     * of the XiveTCTX is set to print out the OS CAM line under the
+     * QEMU monitor.
+     *
+     * This breaks the migration on a TCG guest (or on KVM with
+     * kernel_irqchip=off) because the matching algorithm of the
+     * presenter relies on the OS CAM value. Fix with an extra reset
+     * of the thread contexts to restore the expected value.
+     */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        /* (TCG) Set the OS CAM line of the thread interrupt context. */
+        spapr_xive_set_tctx_os_cam(spapr_cpu_state(cpu)->tctx);
+    }
+    return 0;
 }
 
 static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
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
#20 lore raw

Re: [Qemu-devel] [PATCH v2 07/13] spapr/xive: fix migration of the XiveTCTX under TCG

Quoted 10 lines · David Gibson
Quoted 9 lines · Cédric Le Goater

When the thread interrupt management state is retrieved from the KVM VCPU, word2 is saved under the QEMU XIVE thread context to print out the OS CAM line under the QEMU monitor.

This breaks the migration of a TCG guest (and with KVM when kernel_irqchip=off) because the matching algorithm of the presenter relies on the OS CAM value. Fix with an extra reset of the thread contexts to restore the expected value.

Quoted 4 lines
Quoted 1 line

As noted elsewhere, I'm not sure this is the right approach to fixing this. In any case this can be folded into the previous patch.

I have proposed an alternative in a response to :

[PATCH v2 04/13] spapr/xive: add state synchronization with KVM

C.

Quoted 43 lines
Quoted 42 lines
---
 hw/ppc/spapr_irq.c | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)
hw/ppc/spapr_irq.c +25 -1
index 12ecca6264f3..3176098b9f7c 100644
@@ -356,7 +356,31 @@ static void spapr_irq_cpu_intc_create_xive(sPAPRMachineState *spapr,
 
 static int spapr_irq_post_load_xive(sPAPRMachineState *spapr, int version_id)
 {
-    return spapr_xive_post_load(spapr->xive, version_id);
+    CPUState *cs;
+    int ret;
+
+    ret = spapr_xive_post_load(spapr->xive, version_id);
+    if (ret) {
+        return ret;
+    }
+
+    /*
+     * When the states are collected from the KVM XIVE device, word2
+     * of the XiveTCTX is set to print out the OS CAM line under the
+     * QEMU monitor.
+     *
+     * This breaks the migration on a TCG guest (or on KVM with
+     * kernel_irqchip=off) because the matching algorithm of the
+     * presenter relies on the OS CAM value. Fix with an extra reset
+     * of the thread contexts to restore the expected value.
+     */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        /* (TCG) Set the OS CAM line of the thread interrupt context. */
+        spapr_xive_set_tctx_os_cam(spapr_cpu_state(cpu)->tctx);
+    }
+    return 0;
 }
 
 static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
#21 lore raw

PATCH v2 8/13 spapr/rtas: modify spapr_rtas_register() to remove RTAS handlers

In reply to Cédric Le Goater

Removing RTAS handlers will become necessary when the new pseries machine supporting multiple interrupt mode is introduced.

---
 include/hw/ppc/spapr.h | 4 ++++
 hw/ppc/spapr_rtas.c    | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)
include/hw/ppc/spapr.h +4 -0
index 59073a757900..dd346d921428 100644
@@ -660,6 +660,10 @@ typedef void (*spapr_rtas_fn)(PowerPCCPU *cpu, sPAPRMachineState *sm,
                               uint32_t nargs, target_ulong args,
                               uint32_t nret, target_ulong rets);
 void spapr_rtas_register(int token, const char *name, spapr_rtas_fn fn);
+static inline void spapr_rtas_unregister(int token)
+{
+    spapr_rtas_register(token, NULL, NULL);
+}
 target_ulong spapr_rtas_call(PowerPCCPU *cpu, sPAPRMachineState *sm,
                              uint32_t token, uint32_t nargs, target_ulong args,
                              uint32_t nret, target_ulong rets);
hw/ppc/spapr_rtas.c +1 -1
index 7a2cb786a36a..d09c5f463d22 100644
@@ -404,7 +404,7 @@ void spapr_rtas_register(int token, const char *name, spapr_rtas_fn fn)
 
     token -= RTAS_TOKEN_BASE;
 
-    assert(!rtas_table[token].name);
+    assert(!name || !rtas_table[token].name);
 
     rtas_table[token].name = name;
     rtas_table[token].fn = fn;
Signature · 1 line
2.20.1
#22 lore raw

PATCH v2 9/13 sysbus: add a sysbus_mmio_unmap() helper

In reply to Cédric Le Goater

This will be used to remove the MMIO regions of the POWER9 XIVE interrupt controller when the sPAPR machine is reseted.

---
 include/hw/sysbus.h |  1 +
 hw/core/sysbus.c    | 10 ++++++++++
 2 files changed, 11 insertions(+)
include/hw/sysbus.h +1 -0
index 1aedcf05c92b..4c668fbbdc60 100644
@@ -89,6 +89,7 @@ qemu_irq sysbus_get_connected_irq(SysBusDevice *dev, int n);
 void sysbus_mmio_map(SysBusDevice *dev, int n, hwaddr addr);
 void sysbus_mmio_map_overlap(SysBusDevice *dev, int n, hwaddr addr,
                              int priority);
+void sysbus_mmio_unmap(SysBusDevice *dev, int n);
 void sysbus_add_io(SysBusDevice *dev, hwaddr addr,
                    MemoryRegion *mem);
 MemoryRegion *sysbus_address_space(SysBusDevice *dev);
hw/core/sysbus.c +10 -0
index 9f9edbcab96f..f90d87b058c3 100644
@@ -153,6 +153,16 @@ static void sysbus_mmio_map_common(SysBusDevice *dev, int n, hwaddr addr,
     }
 }
 
+void sysbus_mmio_unmap(SysBusDevice *dev, int n)
+{
+    assert(n >= 0 && n < dev->num_mmio);
+
+    if (dev->mmio[n].addr != (hwaddr)-1) {
+        memory_region_del_subregion(get_system_memory(), dev->mmio[n].memory);
+        dev->mmio[n].addr = (hwaddr)-1;
+    }
+}
+
 void sysbus_mmio_map(SysBusDevice *dev, int n, hwaddr addr)
 {
     sysbus_mmio_map_common(dev, n, addr, false, 0);
Signature · 1 line
2.20.1
#23 lore raw

PATCH v2 10/13 spapr: introduce routines to delete the KVM IRQ device

In reply to Cédric Le Goater

If a new interrupt mode is chosen by CAS, the machine generates a reset to reconfigure. At this point, the connection with the previous KVM device needs to be closed and a new connection needs to opened with the KVM device operating the chosen interrupt mode.

New routines are introduced to destroy the XICS and the XIVE KVM devices. They make use of a new KVM device ioctl which destroys the device and also disconnects the IRQ presenters from the vCPUs.

---
 include/hw/ppc/spapr_xive.h |  1 +
 include/hw/ppc/xics_spapr.h |  1 +
 hw/intc/spapr_xive_kvm.c    | 60 +++++++++++++++++++++++++++++++++++++
 hw/intc/xics_kvm.c          | 56 ++++++++++++++++++++++++++++++++++
 4 files changed, 118 insertions(+)
include/hw/ppc/spapr_xive.h +1 -0
index 22d70650b51f..a7c4c275a747 100644
@@ -71,6 +71,7 @@ int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
  * KVM XIVE device helpers
  */
 void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_disconnect(sPAPRXive *xive, Error **errp);
 void kvmppc_xive_reset(sPAPRXive *xive, Error **errp);
 void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
                                    Error **errp);
include/hw/ppc/xics_spapr.h +1 -0
index b8d924baf437..bddf09821cb0 100644
@@ -34,6 +34,7 @@
 void spapr_dt_xics(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
                    uint32_t phandle);
 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp);
+int xics_kvm_disconnect(sPAPRMachineState *spapr, Error **errp);
 void xics_spapr_init(sPAPRMachineState *spapr);
 
 #endif /* XICS_SPAPR_H */
hw/intc/spapr_xive_kvm.c +60 -0

69 lines not on this page · raw message

hw/intc/xics_kvm.c +56 -0

65 lines not on this page · raw message

Signature · 1 line
2.20.1
Reviewed-by on v2 10/13
#24 lore raw

Re: [Qemu-devel] [PATCH v2 10/13] spapr: introduce routines to delete the KVM IRQ device

Quoted 10 lines · Cédric Le Goater

If a new interrupt mode is chosen by CAS, the machine generates a reset to reconfigure. At this point, the connection with the previous KVM device needs to be closed and a new connection needs to opened with the KVM device operating the chosen interrupt mode.

New routines are introduced to destroy the XICS and the XIVE KVM devices. They make use of a new KVM device ioctl which destroys the device and also disconnects the IRQ presenters from the vCPUs.

Ugly, but necessary

Quoted 177 lines
---
 include/hw/ppc/spapr_xive.h |  1 +
 include/hw/ppc/xics_spapr.h |  1 +
 hw/intc/spapr_xive_kvm.c    | 60 +++++++++++++++++++++++++++++++++++++
 hw/intc/xics_kvm.c          | 56 ++++++++++++++++++++++++++++++++++
 4 files changed, 118 insertions(+)
include/hw/ppc/spapr_xive.h +1 -0
index 22d70650b51f..a7c4c275a747 100644
@@ -71,6 +71,7 @@ int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
  * KVM XIVE device helpers
  */
 void kvmppc_xive_connect(sPAPRXive *xive, Error **errp);
+void kvmppc_xive_disconnect(sPAPRXive *xive, Error **errp);
 void kvmppc_xive_reset(sPAPRXive *xive, Error **errp);
 void kvmppc_xive_set_source_config(sPAPRXive *xive, uint32_t lisn, XiveEAS *eas,
                                    Error **errp);
include/hw/ppc/xics_spapr.h +1 -0
index b8d924baf437..bddf09821cb0 100644
@@ -34,6 +34,7 @@
 void spapr_dt_xics(sPAPRMachineState *spapr, uint32_t nr_servers, void *fdt,
                    uint32_t phandle);
 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp);
+int xics_kvm_disconnect(sPAPRMachineState *spapr, Error **errp);
 void xics_spapr_init(sPAPRMachineState *spapr);
 
 #endif /* XICS_SPAPR_H */
hw/intc/spapr_xive_kvm.c +60 -0
index 119fd59fc9ae..e31035c90260 100644
@@ -58,6 +58,16 @@ static void kvm_cpu_enable(CPUState *cs)
     QLIST_INSERT_HEAD(&kvm_enabled_cpus, enabled_cpu, node);
 }
 
+static void kvm_cpu_disable_all(void)
+{
+    KVMEnabledCPU *enabled_cpu, *next;
+
+    QLIST_FOREACH_SAFE(enabled_cpu, &kvm_enabled_cpus, node, next) {
+        QLIST_REMOVE(enabled_cpu, node);
+        g_free(enabled_cpu);
+    }
+}
+
 /*
  * XIVE Thread Interrupt Management context (KVM)
  */
@@ -674,3 +684,53 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
     /* Map all regions */
     spapr_xive_map_mmio(xive);
 }
+
+void kvmppc_xive_disconnect(sPAPRXive *xive, Error **errp)
+{
+    XiveSource *xsrc;
+    struct kvm_destroy_device xive_destroy_device;
+    size_t esb_len;
+    int rc;
+
+    /* The KVM XIVE device is not in use */
+    if (!xive || xive->fd == -1) {
+        return;
+    }
+
+    if (!kvmppc_has_cap_xive()) {
+        error_setg(errp, "IRQ_XIVE capability must be present for KVM");
+        return;
+    }
+
+    /* Clear the KVM mapping */
+    xsrc = &xive->source;
+    esb_len = (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
+
+    sysbus_mmio_unmap(SYS_BUS_DEVICE(xive), 0);
+    munmap(xsrc->esb_mmap, esb_len);
+
+    sysbus_mmio_unmap(SYS_BUS_DEVICE(xive), 1);
+
+    sysbus_mmio_unmap(SYS_BUS_DEVICE(xive), 2);
+    munmap(xive->tm_mmap, 4ull << TM_SHIFT);
+
+    /* Destroy the KVM device. This also clears the VCPU presenters */
+    xive_destroy_device.fd = xive->fd;
+    xive_destroy_device.flags = 0;
+    rc = kvm_vm_ioctl(kvm_state, KVM_DESTROY_DEVICE, &xive_destroy_device);
+    if (rc < 0) {
+        error_setg_errno(errp, -rc, "Error on KVM_DESTROY_DEVICE for XIVE");
+    }
+    close(xive->fd);
+    xive->fd = -1;
+
+    kvm_kernel_irqchip = false;
+    kvm_msi_via_irqfd_allowed = false;
+    kvm_gsi_direct_mapping = false;
+
+    /* Clear the local list of presenter (hotplug) */
+    kvm_cpu_disable_all();
+
+    /* VM Change state handler is not needed anymore */
+    qemu_del_vm_change_state_handler(xive->change);
+}
hw/intc/xics_kvm.c +56 -0
index c6e1b630a404..373de3155f6b 100644
@@ -51,6 +51,16 @@ typedef struct KVMEnabledICP {
 static QLIST_HEAD(, KVMEnabledICP)
     kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
 
+static void kvm_disable_icps(void)
+{
+    KVMEnabledICP *enabled_icp, *next;
+
+    QLIST_FOREACH_SAFE(enabled_icp, &kvm_enabled_icps, node, next) {
+        QLIST_REMOVE(enabled_icp, node);
+        g_free(enabled_icp);
+    }
+}
+
 /*
  * ICP-KVM
  */
@@ -360,3 +370,49 @@ fail:
     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
     return -1;
 }
+
+int xics_kvm_disconnect(sPAPRMachineState *spapr, Error **errp)
+{
+    int rc;
+    struct kvm_destroy_device xics_destroy_device = {
+        .fd = kernel_xics_fd,
+        .flags = 0,
+    };
+
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return 0;
+    }
+
+    if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
+        error_setg(errp,
+                   "KVM and IRQ_XICS capability must be present for KVM XICS device");
+        return -1;
+    }
+
+    rc = kvm_vm_ioctl(kvm_state, KVM_DESTROY_DEVICE, &xics_destroy_device);
+    if (rc < 0) {
+        error_setg_errno(errp, -rc, "Error on KVM_DESTROY_DEVICE for XICS");
+    }
+    close(kernel_xics_fd);
+    kernel_xics_fd = -1;
+
+    spapr_rtas_unregister(RTAS_IBM_SET_XIVE);
+    spapr_rtas_unregister(RTAS_IBM_GET_XIVE);
+    spapr_rtas_unregister(RTAS_IBM_INT_OFF);
+    spapr_rtas_unregister(RTAS_IBM_INT_ON);
+
+    kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
+    kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
+    kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
+    kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
+
+    kvm_kernel_irqchip = false;
+    kvm_msi_via_irqfd_allowed = false;
+    kvm_gsi_direct_mapping = false;
+
+    /* Clear the presenter from the VCPUs */
+    kvm_disable_icps();
+
+    return rc;
+}
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
#25 lore raw

PATCH v2 11/13 spapr: check for the activation of the KVM IRQ device

In reply to Cédric Le Goater

The activation of the KVM IRQ device depends on the interrupt mode chosen at CAS time by the machine and some methods used at reset or by the migration need to be protected.

---
 hw/intc/spapr_xive_kvm.c | 28 ++++++++++++++++++++++++++++
 hw/intc/xics_kvm.c       | 26 +++++++++++++++++++++++++-
 2 files changed, 53 insertions(+), 1 deletion(-)
hw/intc/spapr_xive_kvm.c +28 -0
index e31035c90260..cd81cdb23a5e 100644
@@ -96,9 +96,15 @@ static void kvmppc_xive_cpu_set_state(XiveTCTX *tctx, Error **errp)
 
 void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp)
 {
+    sPAPRXive *xive = SPAPR_MACHINE(qdev_get_machine())->xive;
     uint64_t state[4] = { 0 };
     int ret;
 
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return;
+    }
+
     ret = kvm_get_one_reg(tctx->cs, KVM_REG_PPC_NVT_STATE, state);
     if (ret != 0) {
         error_setg_errno(errp, errno,
@@ -152,6 +158,11 @@ void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
     unsigned long vcpu_id;
     int ret;
 
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return;
+    }
+
     /* Check if CPU was hot unplugged and replugged. */
     if (kvm_cpu_is_enabled(tctx->cs)) {
         return;
@@ -330,9 +341,13 @@ static void kvmppc_xive_source_get_state(XiveSource *xsrc)
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 {
     XiveSource *xsrc = opaque;
+    sPAPRXive *xive = SPAPR_XIVE(xsrc->xive);
     struct kvm_irq_level args;
     int rc;
 
+    /* The KVM XIVE device should be in use */
+    assert(xive->fd != -1);
+
     args.irq = srcno;
     if (!xive_source_irq_is_lsi(xsrc, srcno)) {
         if (!val) {
@@ -519,6 +534,11 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
 
 void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
 {
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return;
+    }
+
     /*
      * When the VM is stopped, the sources are masked and the previous
      * state is saved in anticipation of a migration. We should not
@@ -544,6 +564,11 @@ int kvmppc_xive_pre_save(sPAPRXive *xive)
 {
     Error *local_err = NULL;
 
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return 0;
+    }
+
     /* EAT: there is no extra state to query from KVM */
 
     /* ENDT */
@@ -568,6 +593,9 @@ int kvmppc_xive_post_load(sPAPRXive *xive, int version_id)
     CPUState *cs;
     int i;
 
+    /* The KVM XIVE device should be in use */
+    assert(xive->fd != -1);
+
     /* Restore the ENDT first. The targetting depends on it. */
     for (i = 0; i < xive->nr_ends; i++) {
         kvmppc_xive_set_queue_config(xive, SPAPR_XIVE_BLOCK_ID, i,
hw/intc/xics_kvm.c +25 -1
index 373de3155f6b..9855316e4831 100644
@@ -69,6 +69,11 @@ void icp_get_kvm_state(ICPState *icp)
     uint64_t state;
     int ret;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return;
+    }
+
     /* ICP for this CPU thread is not in use, exiting */
     if (!icp->cs) {
         return;
@@ -105,6 +110,11 @@ int icp_set_kvm_state(ICPState *icp)
     uint64_t state;
     int ret;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return 0;
+    }
+
     /* ICP for this CPU thread is not in use, exiting */
     if (!icp->cs) {
         return 0;
@@ -133,8 +143,9 @@ void icp_kvm_realize(DeviceState *dev, Error **errp)
     unsigned long vcpu_id;
     int ret;
 
+    /* The KVM XICS device is not in use */
     if (kernel_xics_fd == -1) {
-        abort();
+        return;
     }
 
     cs = icp->cs;
@@ -170,6 +181,11 @@ void ics_get_kvm_state(ICSState *ics)
     uint64_t state;
     int i;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return;
+    }
+
     for (i = 0; i < ics->nr_irqs; i++) {
         ICSIRQState *irq = &ics->irqs[i];
 
@@ -269,6 +285,11 @@ int ics_set_kvm_state(ICSState *ics)
 {
     int i;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return 0;
+    }
+
     for (i = 0; i < ics->nr_irqs; i++) {
         int ret;
 
@@ -286,6 +307,9 @@ void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
     struct kvm_irq_level args;
     int rc;
 
+    /* The KVM XICS device should be in use */
+    assert(kernel_xics_fd != -1);
+
     args.irq = srcno + ics->offset;
     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
         if (!val) {
Signature · 1 line
2.20.1
Reviewed-by on v2 11/13
#26 lore raw

Re: [Qemu-devel] [PATCH v2 11/13] spapr: check for the activation of the KVM IRQ device

Quoted 5 lines · Cédric Le Goater

The activation of the KVM IRQ device depends on the interrupt mode chosen at CAS time by the machine and some methods used at reset or by the migration need to be protected.

Quoted 158 lines
---
 hw/intc/spapr_xive_kvm.c | 28 ++++++++++++++++++++++++++++
 hw/intc/xics_kvm.c       | 26 +++++++++++++++++++++++++-
 2 files changed, 53 insertions(+), 1 deletion(-)
hw/intc/spapr_xive_kvm.c +28 -0
index e31035c90260..cd81cdb23a5e 100644
@@ -96,9 +96,15 @@ static void kvmppc_xive_cpu_set_state(XiveTCTX *tctx, Error **errp)
 
 void kvmppc_xive_cpu_get_state(XiveTCTX *tctx, Error **errp)
 {
+    sPAPRXive *xive = SPAPR_MACHINE(qdev_get_machine())->xive;
     uint64_t state[4] = { 0 };
     int ret;
 
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return;
+    }
+
     ret = kvm_get_one_reg(tctx->cs, KVM_REG_PPC_NVT_STATE, state);
     if (ret != 0) {
         error_setg_errno(errp, errno,
@@ -152,6 +158,11 @@ void kvmppc_xive_cpu_connect(XiveTCTX *tctx, Error **errp)
     unsigned long vcpu_id;
     int ret;
 
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return;
+    }
+
     /* Check if CPU was hot unplugged and replugged. */
     if (kvm_cpu_is_enabled(tctx->cs)) {
         return;
@@ -330,9 +341,13 @@ static void kvmppc_xive_source_get_state(XiveSource *xsrc)
 void kvmppc_xive_source_set_irq(void *opaque, int srcno, int val)
 {
     XiveSource *xsrc = opaque;
+    sPAPRXive *xive = SPAPR_XIVE(xsrc->xive);
     struct kvm_irq_level args;
     int rc;
 
+    /* The KVM XIVE device should be in use */
+    assert(xive->fd != -1);
+
     args.irq = srcno;
     if (!xive_source_irq_is_lsi(xsrc, srcno)) {
         if (!val) {
@@ -519,6 +534,11 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
 
 void kvmppc_xive_synchronize_state(sPAPRXive *xive, Error **errp)
 {
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return;
+    }
+
     /*
      * When the VM is stopped, the sources are masked and the previous
      * state is saved in anticipation of a migration. We should not
@@ -544,6 +564,11 @@ int kvmppc_xive_pre_save(sPAPRXive *xive)
 {
     Error *local_err = NULL;
 
+    /* The KVM XIVE device is not in use */
+    if (xive->fd == -1) {
+        return 0;
+    }
+
     /* EAT: there is no extra state to query from KVM */
 
     /* ENDT */
@@ -568,6 +593,9 @@ int kvmppc_xive_post_load(sPAPRXive *xive, int version_id)
     CPUState *cs;
     int i;
 
+    /* The KVM XIVE device should be in use */
+    assert(xive->fd != -1);
+
     /* Restore the ENDT first. The targetting depends on it. */
     for (i = 0; i < xive->nr_ends; i++) {
         kvmppc_xive_set_queue_config(xive, SPAPR_XIVE_BLOCK_ID, i,
hw/intc/xics_kvm.c +25 -1
index 373de3155f6b..9855316e4831 100644
@@ -69,6 +69,11 @@ void icp_get_kvm_state(ICPState *icp)
     uint64_t state;
     int ret;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return;
+    }
+
     /* ICP for this CPU thread is not in use, exiting */
     if (!icp->cs) {
         return;
@@ -105,6 +110,11 @@ int icp_set_kvm_state(ICPState *icp)
     uint64_t state;
     int ret;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return 0;
+    }
+
     /* ICP for this CPU thread is not in use, exiting */
     if (!icp->cs) {
         return 0;
@@ -133,8 +143,9 @@ void icp_kvm_realize(DeviceState *dev, Error **errp)
     unsigned long vcpu_id;
     int ret;
 
+    /* The KVM XICS device is not in use */
     if (kernel_xics_fd == -1) {
-        abort();
+        return;
     }
 
     cs = icp->cs;
@@ -170,6 +181,11 @@ void ics_get_kvm_state(ICSState *ics)
     uint64_t state;
     int i;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return;
+    }
+
     for (i = 0; i < ics->nr_irqs; i++) {
         ICSIRQState *irq = &ics->irqs[i];
 
@@ -269,6 +285,11 @@ int ics_set_kvm_state(ICSState *ics)
 {
     int i;
 
+    /* The KVM XICS device is not in use */
+    if (kernel_xics_fd == -1) {
+        return 0;
+    }
+
     for (i = 0; i < ics->nr_irqs; i++) {
         int ret;
 
@@ -286,6 +307,9 @@ void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
     struct kvm_irq_level args;
     int rc;
 
+    /* The KVM XICS device should be in use */
+    assert(kernel_xics_fd != -1);
+
     args.irq = srcno + ics->offset;
     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
         if (!val) {
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
#27 lore raw

PATCH v2 12/13 spapr: add KVM support to the 'dual' machine

In reply to Cédric Le Goater

The interrupt mode is chosen by the CAS negotiation process and activated after a reset to take into account the required changes in the machine. This brings new constraints on how the associated KVM IRQ device is initialized.

Currently, each model takes care of the initialization of the KVM device in their realize method but this is not possible anymore as the initialization needs to be done globaly when the interrupt mode is known, i.e. when machine is reseted. It also means that we need a way to delete a KVM device when another mode is chosen.

Also, to support migration, the QEMU objects holding the state to transfer should always be available but not necessarily activated.

The overall approach of this proposal is to initialize both interrupt mode at the QEMU level and keep the IRQ number space in sync to allow switching from one mode to another. For the KVM side of things, the whole initialization of the KVM device, sources and presenters, is grouped in a single routine. The XICS and XIVE sPAPR IRQ reset handlers are modified accordingly to handle the init and the delete sequences of the KVM device.

---
 include/hw/ppc/spapr_xive.h |  1 +
 hw/intc/spapr_xive.c        | 19 +++++++-
 hw/intc/spapr_xive_kvm.c    | 27 +++++++++++
 hw/intc/xics_kvm.c          | 26 ++++++++++
 hw/intc/xive.c              |  4 --
 hw/ppc/spapr_irq.c          | 97 ++++++++++++++++++++++++++++---------
 6 files changed, 145 insertions(+), 29 deletions(-)
include/hw/ppc/spapr_xive.h +1 -0
index a7c4c275a747..a1593ac2fcf0 100644
@@ -66,6 +66,7 @@ void spapr_xive_map_mmio(sPAPRXive *xive);
 
 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
                              uint32_t *out_server, uint8_t *out_prio);
+void spapr_xive_late_realize(sPAPRXive *xive, Error **errp);
 
 /*
  * KVM XIVE device helpers
hw/intc/spapr_xive.c +18 -1
index 21fe5e1aa39f..b0cbc2fe21ee 100644
@@ -278,7 +278,6 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
     XiveSource *xsrc = &xive->source;
     XiveENDSource *end_xsrc = &xive->end_source;
     Error *local_err = NULL;
-    MachineState *machine = MACHINE(qdev_get_machine());
 
     if (!xive->nr_irqs) {
         error_setg(errp, "Number of interrupt needs to be greater 0");
@@ -329,6 +328,15 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
 
     qemu_register_reset(spapr_xive_reset, dev);
+}
+
+void spapr_xive_late_realize(sPAPRXive *xive, Error **errp)
+{
+    Error *local_err = NULL;
+    MachineState *machine = MACHINE(qdev_get_machine());
+    XiveSource *xsrc = &xive->source;
+    XiveENDSource *end_xsrc = &xive->end_source;
+    static bool once;
 
     if (kvm_enabled() && machine_kernel_irqchip_allowed(machine)) {
         kvmppc_xive_connect(xive, &local_err);
@@ -351,6 +359,15 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
         error_report_err(local_err);
     }
 
+    /*
+     * TODO: Emulated mode can only be initialized once. Should we
+     * store the information under the device model for later usage ?
+     */
+    if (once) {
+        return;
+    }
+    once = true;
+
     /* TIMA initialization */
     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
                           "xive.tima", 4ull << TM_SHIFT);
hw/intc/spapr_xive_kvm.c +27 -0
index cd81cdb23a5e..99a829fb3f60 100644
@@ -657,6 +657,15 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
     Error *local_err = NULL;
     size_t esb_len = (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
     size_t tima_len = 4ull << TM_SHIFT;
+    CPUState *cs;
+
+    /*
+     * The KVM XIVE device already in use. This is the case when
+     * rebooting XIVE -> XIVE
+     */
+    if (xive->fd != -1) {
+        return;
+    }
 
     if (!kvmppc_has_cap_xive()) {
         error_setg(errp, "IRQ_XIVE capability must be present for KVM");
@@ -705,6 +714,24 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
     xive->change = qemu_add_vm_change_state_handler(
         kvmppc_xive_change_state_handler, xive);
 
+    /* Connect the presenters to the initial VCPUs of the machine */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        kvmppc_xive_cpu_connect(spapr_cpu_state(cpu)->tctx, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+
+    /* Update the KVM sources */
+    kvmppc_xive_source_reset(xsrc, &local_err);
+    if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+    }
+
     kvm_kernel_irqchip = true;
     kvm_msi_via_irqfd_allowed = true;
     kvm_gsi_direct_mapping = true;
hw/intc/xics_kvm.c +26 -0
index 9855316e4831..8ffd4c7a36f8 100644
@@ -33,6 +33,7 @@
 #include "trace.h"
 #include "sysemu/kvm.h"
 #include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_cpu_core.h"
 #include "hw/ppc/xics.h"
 #include "hw/ppc/xics_spapr.h"
 #include "kvm_ppc.h"
@@ -337,6 +338,16 @@ static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 {
     int rc;
+    CPUState *cs;
+    Error *local_err = NULL;
+
+    /*
+     * The KVM XICS device already in use. This is the case when
+     * rebooting XICS -> XICS
+     */
+    if (kernel_xics_fd != -1) {
+        return 0;
+    }
 
     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
         error_setg(errp,
@@ -385,6 +396,21 @@ int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
     kvm_msi_via_irqfd_allowed = true;
     kvm_gsi_direct_mapping = true;
 
+    /* Connect the presenters to the initial VCPUs of the machine */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            goto fail;
+        }
+        icp_set_kvm_state(spapr_cpu_state(cpu)->icp);
+    }
+
+    /* Update the KVM sources */
+    ics_set_kvm_state(spapr->ics);
+
     return 0;
 
 fail:
hw/intc/xive.c +0 -4
index 1f8e923ca654..715d5a7e65ed 100644
@@ -929,10 +929,6 @@ static void xive_source_reset(void *dev)
 
     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
-
-    if (kvm_irqchip_in_kernel()) {
-        kvmppc_xive_source_reset(xsrc, &error_fatal);
-    }
 }
 
 static void xive_source_realize(DeviceState *dev, Error **errp)
hw/ppc/spapr_irq.c +73 -24

165 lines not on this page · raw message

Signature · 1 line
2.20.1
#28 lore raw

Re: [Qemu-devel] [PATCH v2 12/13] spapr: add KVM support to the 'dual' machine

Quoted 74 lines · Cédric Le Goater

The interrupt mode is chosen by the CAS negotiation process and activated after a reset to take into account the required changes in the machine. This brings new constraints on how the associated KVM IRQ device is initialized.

Currently, each model takes care of the initialization of the KVM device in their realize method but this is not possible anymore as the initialization needs to be done globaly when the interrupt mode is known, i.e. when machine is reseted. It also means that we need a way to delete a KVM device when another mode is chosen.

Also, to support migration, the QEMU objects holding the state to transfer should always be available but not necessarily activated.

The overall approach of this proposal is to initialize both interrupt mode at the QEMU level and keep the IRQ number space in sync to allow switching from one mode to another. For the KVM side of things, the whole initialization of the KVM device, sources and presenters, is grouped in a single routine. The XICS and XIVE sPAPR IRQ reset handlers are modified accordingly to handle the init and the delete sequences of the KVM device.

---
 include/hw/ppc/spapr_xive.h |  1 +
 hw/intc/spapr_xive.c        | 19 +++++++-
 hw/intc/spapr_xive_kvm.c    | 27 +++++++++++
 hw/intc/xics_kvm.c          | 26 ++++++++++
 hw/intc/xive.c              |  4 --
 hw/ppc/spapr_irq.c          | 97 ++++++++++++++++++++++++++++---------
 6 files changed, 145 insertions(+), 29 deletions(-)
include/hw/ppc/spapr_xive.h +1 -0
index a7c4c275a747..a1593ac2fcf0 100644
@@ -66,6 +66,7 @@ void spapr_xive_map_mmio(sPAPRXive *xive);
 
 int spapr_xive_end_to_target(uint8_t end_blk, uint32_t end_idx,
                              uint32_t *out_server, uint8_t *out_prio);
+void spapr_xive_late_realize(sPAPRXive *xive, Error **errp);
 
 /*
  * KVM XIVE device helpers
hw/intc/spapr_xive.c +9 -1
index 21fe5e1aa39f..b0cbc2fe21ee 100644
@@ -278,7 +278,6 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
     XiveSource *xsrc = &xive->source;
     XiveENDSource *end_xsrc = &xive->end_source;
     Error *local_err = NULL;
-    MachineState *machine = MACHINE(qdev_get_machine());
 
     if (!xive->nr_irqs) {
         error_setg(errp, "Number of interrupt needs to be greater 0");
@@ -329,6 +328,15 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
                            xive->tm_base + XIVE_TM_USER_PAGE * (1 << TM_SHIFT));
 
     qemu_register_reset(spapr_xive_reset, dev);
+}
+
+void spapr_xive_late_realize(sPAPRXive *xive, Error **errp)
+{
+    Error *local_err = NULL;
+    MachineState *machine = MACHINE(qdev_get_machine());
+    XiveSource *xsrc = &xive->source;
+    XiveENDSource *end_xsrc = &xive->end_source;
+    static bool once;
 
     if (kvm_enabled() && machine_kernel_irqchip_allowed(machine)) {
         kvmppc_xive_connect(xive, &local_err);
@@ -351,6 +359,15 @@ static void spapr_xive_realize(DeviceState *dev, Error **errp)
         error_report_err(local_err);

hw/intc/spapr_xive.c static void spapr_xive_realize(DeviceState *dev, Error **errp) · line 369

Quoted 10 lines
hw/intc/spapr_xive.c +8 -0
     }
 
+    /*
+     * TODO: Emulated mode can only be initialized once. Should we
+     * store the information under the device model for later usage ?
+     */
+    if (once) {
+        return;
+    }
+    once = true;

Urgh. static locals are a bad smell. I think at least this flag should go into the instance structure (if we can't deduce it from something else in there).

Quoted 9 lines
hw/intc/spapr_xive.c +1 -0
+
     /* TIMA initialization */
     memory_region_init_io(&xive->tm_mmio, OBJECT(xive), &xive_tm_ops, xive,
                           "xive.tima", 4ull << TM_SHIFT);
hw/intc/spapr_xive_kvm.c +0 -0
index cd81cdb23a5e..99a829fb3f60 100644

hw/intc/spapr_xive_kvm.c void kvmppc_xive_connect(sPAPRXive *xive, Error **errp) · line 664

Quoted 8 lines
hw/intc/spapr_xive_kvm.c +5 -0
@@ -657,6 +657,15 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
     Error *local_err = NULL;
     size_t esb_len = (1ull << xsrc->esb_shift) * xsrc->nr_irqs;
     size_t tima_len = 4ull << TM_SHIFT;
+    CPUState *cs;
+
+    /*
+     * The KVM XIVE device already in use. This is the case when
+     * rebooting XIVE -> XIVE

I take it from this we're leaving the XIVE KVM device initialized if we don't change to XICS during CAS. Would it make things simpler if we always removed both the XICS and XIVE KVM devices at reset and recreated the one we need at CAS?

Quoted 133 lines
hw/intc/spapr_xive_kvm.c +22 -0
+     */
+    if (xive->fd != -1) {
+        return;
+    }
 
     if (!kvmppc_has_cap_xive()) {
         error_setg(errp, "IRQ_XIVE capability must be present for KVM");
@@ -705,6 +714,24 @@ void kvmppc_xive_connect(sPAPRXive *xive, Error **errp)
     xive->change = qemu_add_vm_change_state_handler(
         kvmppc_xive_change_state_handler, xive);
 
+    /* Connect the presenters to the initial VCPUs of the machine */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        kvmppc_xive_cpu_connect(spapr_cpu_state(cpu)->tctx, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+        }
+    }
+
+    /* Update the KVM sources */
+    kvmppc_xive_source_reset(xsrc, &local_err);
+    if (local_err) {
+            error_propagate(errp, local_err);
+            return;
+    }
+
     kvm_kernel_irqchip = true;
     kvm_msi_via_irqfd_allowed = true;
     kvm_gsi_direct_mapping = true;
hw/intc/xics_kvm.c +26 -0
index 9855316e4831..8ffd4c7a36f8 100644
@@ -33,6 +33,7 @@
 #include "trace.h"
 #include "sysemu/kvm.h"
 #include "hw/ppc/spapr.h"
+#include "hw/ppc/spapr_cpu_core.h"
 #include "hw/ppc/xics.h"
 #include "hw/ppc/xics_spapr.h"
 #include "kvm_ppc.h"
@@ -337,6 +338,16 @@ static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
 {
     int rc;
+    CPUState *cs;
+    Error *local_err = NULL;
+
+    /*
+     * The KVM XICS device already in use. This is the case when
+     * rebooting XICS -> XICS
+     */
+    if (kernel_xics_fd != -1) {
+        return 0;
+    }
 
     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
         error_setg(errp,
@@ -385,6 +396,21 @@ int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
     kvm_msi_via_irqfd_allowed = true;
     kvm_gsi_direct_mapping = true;
 
+    /* Connect the presenters to the initial VCPUs of the machine */
+    CPU_FOREACH(cs) {
+        PowerPCCPU *cpu = POWERPC_CPU(cs);
+
+        icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            goto fail;
+        }
+        icp_set_kvm_state(spapr_cpu_state(cpu)->icp);
+    }
+
+    /* Update the KVM sources */
+    ics_set_kvm_state(spapr->ics);
+
     return 0;
 
 fail:
hw/intc/xive.c +0 -4
index 1f8e923ca654..715d5a7e65ed 100644
@@ -929,10 +929,6 @@ static void xive_source_reset(void *dev)
 
     /* PQs are initialized to 0b01 (Q=1) which corresponds to "ints off" */
     memset(xsrc->status, XIVE_ESB_OFF, xsrc->nr_irqs);
-
-    if (kvm_irqchip_in_kernel()) {
-        kvmppc_xive_source_reset(xsrc, &error_fatal);
-    }
 }
 
 static void xive_source_realize(DeviceState *dev, Error **errp)
hw/ppc/spapr_irq.c +8 -12
index 3176098b9f7c..f8260c14aecd 100644
@@ -92,35 +92,55 @@ error:
     return NULL;
 }
 
-static void spapr_irq_init_xics(sPAPRMachineState *spapr, int nr_irqs,
-                                Error **errp)
+static void spapr_ics_late_realize(sPAPRMachineState *spapr, Error **errp)
 {
     MachineState *machine = MACHINE(spapr);
     Error *local_err = NULL;
-    bool xics_kvm = false;
+    static bool once;
 
-    if (kvm_enabled()) {
-        if (machine_kernel_irqchip_allowed(machine) &&
-            !xics_kvm_init(spapr, &local_err)) {
-            xics_kvm = true;
-        }
-        if (machine_kernel_irqchip_required(machine) && !xics_kvm) {
+    if (kvm_enabled() && machine_kernel_irqchip_allowed(machine)) {
+        xics_kvm_init(spapr, &local_err);
+        if (local_err && machine_kernel_irqchip_required(machine)) {
             error_prepend(&local_err,
                           "kernel_irqchip requested but unavailable: ");
-            goto error;
+            error_propagate(errp, local_err);
+            return;
         }
-        error_free(local_err);
-        local_err = NULL;
+

hw/ppc/spapr_irq.c error: · line 119

Quoted 10 lines
hw/ppc/spapr_irq.c +10 -0
+        if (!local_err) {
+            return;
+        }
+
+        /*
+         * We failed to initialize the XIVE KVM device, fallback to
+         * emulated mode
+         */
+        error_prepend(&local_err, "kernel_irqchip allowed but unavailable: ");
+        error_report_err(local_err);

Should only warn (at most) in this case, since fallback is permitted and should work.

Quoted 84 lines
hw/ppc/spapr_irq.c +33 -12
     }
 
-    if (!xics_kvm) {
-        xics_spapr_init(spapr);
+    /*
+     * TODO: Emulated mode can only be initialized once. Should we
+     * store the information under the device model for later usage ?
+     */
+    if (once) {
+        return;
     }
+    once = true;
 
-    spapr->ics = spapr_ics_create(spapr, nr_irqs, &local_err);
+    xics_spapr_init(spapr);
+}
 
-error:
-    error_propagate(errp, local_err);
+static void spapr_irq_init_xics(sPAPRMachineState *spapr, int nr_irqs,
+                                Error **errp)
+{
+    Error *local_err = NULL;
+
+    spapr->ics = spapr_ics_create(spapr, nr_irqs, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
 }
 
 #define ICS_IRQ_FREE(ics, srcno)   \
@@ -227,7 +247,13 @@ static void spapr_irq_set_irq_xics(void *opaque, int srcno, int val)
 
 static void spapr_irq_reset_xics(sPAPRMachineState *spapr, Error **errp)
 {
-    /* TODO: create the KVM XICS device */
+    Error *local_err = NULL;
+
+    spapr_ics_late_realize(spapr, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
 }
 
 static const char *spapr_irq_get_nodename_xics(sPAPRMachineState *spapr)
@@ -386,6 +412,7 @@ static int spapr_irq_post_load_xive(sPAPRMachineState *spapr, int version_id)
 static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
 {
     CPUState *cs;
+    Error *local_err = NULL;
 
     CPU_FOREACH(cs) {
         PowerPCCPU *cpu = POWERPC_CPU(cs);
@@ -394,6 +421,12 @@ static void spapr_irq_reset_xive(sPAPRMachineState *spapr, Error **errp)
         spapr_xive_set_tctx_os_cam(spapr_cpu_state(cpu)->tctx);
     }
 
+    spapr_xive_late_realize(spapr->xive, &local_err);
+    if (local_err) {
+        error_propagate(errp, local_err);
+        return;
+    }
+
     /* Activate the XIVE MMIOs */
     spapr_xive_mmio_set_enabled(spapr->xive, true);
 }
@@ -462,14 +495,8 @@ static sPAPRIrq *spapr_irq_current(sPAPRMachineState *spapr)
 static void spapr_irq_init_dual(sPAPRMachineState *spapr, int nr_irqs,
                                 Error **errp)
 {
-    MachineState *machine = MACHINE(spapr);
     Error *local_err = NULL;
 
-    if (kvm_enabled() && machine_kernel_irqchip_allowed(machine)) {
-        error_setg(errp, "No KVM support for the 'dual' machine");
-        return;
-    }
-
     spapr_irq_xics.init(spapr, spapr_irq_xics.nr_irqs, &local_err);
     if (local_err) {
         error_propagate(errp, local_err);

hw/ppc/spapr_irq.c static int spapr_irq_post_load_dual(sPAPRMachineState *spapr, int version_id) · line 579

Quoted 5 lines
hw/ppc/spapr_irq.c +2 -0
@@ -548,6 +575,9 @@ static int spapr_irq_post_load_dual(sPAPRMachineState *spapr, int version_id)
      * defaults to XICS at startup.
      */
     if (spapr_ovec_test(spapr->ov5_cas, OV5_XIVE_EXPLOIT)) {
+        if (kvm_irqchip_in_kernel()) {
+            xics_kvm_disconnect(spapr, &error_fatal);

Yeah, this is kinda nasty. I'm wondering if we could make things simpler by always using emulated irqchip until CAS and only setting up kernel irqchip then.

Quoted 43 lines
hw/ppc/spapr_irq.c +20 -0
+        }
         spapr_irq_xive.reset(spapr, &error_fatal);
     }
 
@@ -556,12 +586,30 @@ static int spapr_irq_post_load_dual(sPAPRMachineState *spapr, int version_id)
 
 static void spapr_irq_reset_dual(sPAPRMachineState *spapr, Error **errp)
 {
+    Error *local_err = NULL;
+
     /*
      * Deactivate the XIVE MMIOs. The XIVE backend will reenable them
      * if selected.
      */
     spapr_xive_mmio_set_enabled(spapr->xive, false);
 
+    /* Destroy all KVM devices */
+    if (kvm_irqchip_in_kernel()) {
+        xics_kvm_disconnect(spapr, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            error_prepend(errp, "KVM XICS disconnect failed: ");
+            return;
+        }
+        kvmppc_xive_disconnect(spapr->xive, &local_err);
+        if (local_err) {
+            error_propagate(errp, local_err);
+            error_prepend(errp, "KVM XIVE disconnect failed: ");
+            return;
+        }
+    }
+
     spapr_irq_current(spapr)->reset(spapr, errp);
 }
 
@@ -748,6 +796,7 @@ sPAPRIrq spapr_irq_xics_legacy = {
     .dt_populate = spapr_dt_xics,
     .cpu_intc_create = spapr_irq_cpu_intc_create_xics,
     .post_load   = spapr_irq_post_load_xics,
+    .reset       = spapr_irq_reset_xics,
     .set_irq     = spapr_irq_set_irq_xics,
     .get_nodename = spapr_irq_get_nodename_xics,
 };
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
#29 lore raw

PATCH v2 13/13 spapr/xive: fix device hotplug when VM is stopped

In reply to Cédric Le Goater

Instead of switching off the sources, set their state to PENDING to possibly catch a hotplug event occuring while the VM is stopped. At resume, check the previous state and if an interrupt was queued, generate a trigger.

---
 hw/intc/spapr_xive_kvm.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)
hw/intc/spapr_xive_kvm.c +19 -3
index 99a829fb3f60..64d160babb26 100644
@@ -500,8 +500,16 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
     if (running) {
         for (i = 0; i < xsrc->nr_irqs; i++) {
             uint8_t pq = xive_source_esb_get(xsrc, i);
-            if (xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8)) != 0x1) {
-                error_report("XIVE: IRQ %d has an invalid state", i);
+            uint8_t old_pq;
+
+            old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
+
+            /*
+             * If an interrupt was queued (hotplug event) while VM was
+             * stopped, generate a trigger.
+             */
+            if (pq == XIVE_ESB_RESET && old_pq == XIVE_ESB_QUEUED) {
+                xive_esb_trigger(xsrc, i);
             }
         }
 
@@ -515,7 +523,15 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
      * migration is in progress.
      */
     for (i = 0; i < xsrc->nr_irqs; i++) {
-        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_01);
+        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+
+        /*
+         * PQ is set to PENDING to possibly catch a hotplug event
+         * occuring while the VM is stopped.
+         */
+        if (pq != XIVE_ESB_OFF) {
+            pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_10);
+        }
         xive_source_esb_set(xsrc, i, pq);
     }
 
Signature · 1 line
2.20.1
#30 lore raw

Re: [Qemu-devel] [PATCH v2 13/13] spapr/xive: fix device hotplug when VM is stopped

Quoted 4 lines · Cédric Le Goater

Instead of switching off the sources, set their state to PENDING to possibly catch a hotplug event occuring while the VM is stopped. At resume, check the previous state and if an interrupt was queued, generate a trigger.

First, I think it would be better to fold this fix into the patch introducing the state change handlers.

Second, IIUC this would handle any instance of an irq being triggered while the VM is stopped. Hotplug interrupts is one obvious case of that, but I'm not sure its the only one. VFIO devices could interrupt while the VM is stopped, I think. Maybe even emulated devices depending on how their synchronization with the cpu run state works. There might be other cases. Does that sound right to you?

Quoted 45 lines
---
 hw/intc/spapr_xive_kvm.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)
hw/intc/spapr_xive_kvm.c +19 -3
index 99a829fb3f60..64d160babb26 100644
@@ -500,8 +500,16 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
     if (running) {
         for (i = 0; i < xsrc->nr_irqs; i++) {
             uint8_t pq = xive_source_esb_get(xsrc, i);
-            if (xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8)) != 0x1) {
-                error_report("XIVE: IRQ %d has an invalid state", i);
+            uint8_t old_pq;
+
+            old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
+
+            /*
+             * If an interrupt was queued (hotplug event) while VM was
+             * stopped, generate a trigger.
+             */
+            if (pq == XIVE_ESB_RESET && old_pq == XIVE_ESB_QUEUED) {
+                xive_esb_trigger(xsrc, i);
             }
         }
 
@@ -515,7 +523,15 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
      * migration is in progress.
      */
     for (i = 0; i < xsrc->nr_irqs; i++) {
-        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_01);
+        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+
+        /*
+         * PQ is set to PENDING to possibly catch a hotplug event
+         * occuring while the VM is stopped.
+         */
+        if (pq != XIVE_ESB_OFF) {
+            pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_10);
+        }
         xive_source_esb_set(xsrc, i, pq);
     }
 
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
#31 lore raw

Re: [Qemu-devel] [PATCH v2 13/13] spapr/xive: fix device hotplug when VM is stopped

Quoted 10 lines · David Gibson
Quoted 4 lines · Cédric Le Goater

Instead of switching off the sources, set their state to PENDING to possibly catch a hotplug event occuring while the VM is stopped. At resume, check the previous state and if an interrupt was queued, generate a trigger.

First, I think it would be better to fold this fix into the patch introducing the state change handlers.>
Second, IIUC this would handle any instance of an irq being triggered while the VM is stopped.

yes.

Quoted 2 lines

Hotplug interrupts is one obvious case of that, but I'm not sure its the only one.

Do we really need to support device hotplug when the VM is stopped ? Is that a libvirt requirement ?

Quoted 1 line

VFIO devices could interrupt while the VM is stopped, I think.

If the guest has configured and mapped the IRQs, I would say yes.

Quoted 2 lines

Maybe even emulated devices
depending on how their synchronization with the cpu run state works.

The console is one example.

Quoted 1 line

There might be other cases. Does that sound right to you?

yes.

Supporting interrupts while a VM is stopped seems like a weird test scenario to me. Should we or should we not ?

Thanks,

C.

Quoted 46 lines
Quoted 45 lines
---
 hw/intc/spapr_xive_kvm.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)
hw/intc/spapr_xive_kvm.c +19 -3
index 99a829fb3f60..64d160babb26 100644
@@ -500,8 +500,16 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
     if (running) {
         for (i = 0; i < xsrc->nr_irqs; i++) {
             uint8_t pq = xive_source_esb_get(xsrc, i);
-            if (xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8)) != 0x1) {
-                error_report("XIVE: IRQ %d has an invalid state", i);
+            uint8_t old_pq;
+
+            old_pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_00 + (pq << 8));
+
+            /*
+             * If an interrupt was queued (hotplug event) while VM was
+             * stopped, generate a trigger.
+             */
+            if (pq == XIVE_ESB_RESET && old_pq == XIVE_ESB_QUEUED) {
+                xive_esb_trigger(xsrc, i);
             }
         }
 
@@ -515,7 +523,15 @@ static void kvmppc_xive_change_state_handler(void *opaque, int running,
      * migration is in progress.
      */
     for (i = 0; i < xsrc->nr_irqs; i++) {
-        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_01);
+        uint8_t pq = xive_esb_read(xsrc, i, XIVE_ESB_GET);
+
+        /*
+         * PQ is set to PENDING to possibly catch a hotplug event
+         * occuring while the VM is stopped.
+         */
+        if (pq != XIVE_ESB_OFF) {
+            pq = xive_esb_read(xsrc, i, XIVE_ESB_SET_PQ_10);
+        }
         xive_source_esb_set(xsrc, i, pq);
     }