[Qemu-devel] [PULL 00/21] SCSI, RCU, KVM changes for 2015-02-16

[Qemu-devel] [PULL 00/21] SCSI, RCU, KVM changes for 2015-02-16

[Paolo Bonzini]

The following changes since commit bc5baffa3554e4c0d20c1dbe879aec931866bd69:

  vfio: Fix debug message compile error (2015-02-10 10:25:44 -0700)

are available in the git repository at:

  git://github.com/bonzini/qemu.git tags/for-upstream

for you to fetch changes up to 0dc3f44aca18b1be8b425f3f4feb4b3e8d68de2e:

  Convert ram_list to RCU (2015-02-16 17:31:55 +0100)

----------------------------------------------------------------
- vhost-scsi: add bootindex property
- RCU: fix MemoryRegion lifetime issues in PCI; document the rules;
convert of AddressSpaceDispatch and RAMList
- KVM: add kvm_exit reasons for aarch64

----------------------------------------------------------------
Gonglei (5):
      qdev: support to get a device firmware path directly
      vhost-scsi: add bootindex property
      vhost-scsi: expose the TYPE_FW_PATH_PROVIDER interface
      vhost-scsi: add a property for booting
      vhost-scsi: set the bootable value of channel/target/lun

Mike Day (4):
      rcu: introduce RCU-enabled QLIST
      cosmetic changes preparing for the following patches
      exec: convert ram_list to QLIST
      Convert ram_list to RCU

Paolo Bonzini (11):
      rcu: do not let RCU callbacks pile up indefinitely
      rcu: run RCU callbacks under the BQL
      memory: keep the owner of the AddressSpace alive until do_address_space_destroy
      pcie: remove mmconfig memory leak and wrap mmconfig update with transaction
      pci: split shpc_cleanup and shpc_free
      docs: clarify memory region lifecycle
      exec: introduce cpu_reload_memory_map
      exec: make iotlb RCU-friendly
      exec: RCUify AddressSpaceDispatch
      rcu: add g_free_rcu
      exec: protect mru_block with RCU

Wei Huang (1):
      kvm_stat: Add kvm_exit reasons for aarch64

 arch_init.c                     |  73 +++++++---
 bootdevice.c                    |  31 ++--
 cpu-exec.c                      |  33 +++++
 cpus.c                          |   2 +-
 cputlb.c                        |  13 +-
 docs/memory.txt                 |  74 +++++++---
 docs/rcu.txt                    |  11 +-
 exec.c                          | 295 ++++++++++++++++++++++++++------------
 hw/9pfs/virtio-9p-synth.c       |   2 +-
 hw/core/qdev.c                  |   7 +
 hw/i386/intel_iommu.c           |   3 +
 hw/pci-bridge/pci_bridge_dev.c  |  14 +-
 hw/pci-host/apb.c               |   1 +
 hw/pci/pcie_host.c              |   7 +-
 hw/pci/shpc.c                   |  11 +-
 hw/ppc/spapr_iommu.c            |   1 +
 hw/scsi/vhost-scsi.c            |  35 +++++
 hw/virtio/virtio-pci.c          |   2 +
 include/exec/cpu-all.h          |  13 +-
 include/exec/cputlb.h           |   2 +-
 include/exec/exec-all.h         |   5 +-
 include/hw/pci/shpc.h           |   1 +
 include/hw/qdev-core.h          |   1 +
 include/hw/virtio/vhost-scsi.h  |   5 +
 include/hw/virtio/virtio-scsi.h |   1 +
 include/qemu/queue.h            |  11 --
 include/qemu/rcu.h              |   8 ++
 include/qemu/rcu_queue.h        | 134 ++++++++++++++++++
 include/qom/cpu.h               |   1 +
 memory.c                        |   5 +
 scripts/dump-guest-memory.py    |   8 +-
 scripts/kvm/kvm_stat            |  42 +++++-
 softmmu_template.h              |   4 +-
 tests/Makefile                  |   7 +-
 tests/test-rcu-list.c           | 306 ++++++++++++++++++++++++++++++++++++++++
 util/rcu.c                      |  19 ++-
 36 files changed, 997 insertions(+), 191 deletions(-)
 create mode 100644 include/qemu/rcu_queue.h
 create mode 100644 tests/test-rcu-list.c
-- 
2.3.0

[Qemu-devel] [PULL 01/21] kvm_stat: Add kvm_exit reasons for aarch64

[Paolo Bonzini]

From: Wei Huang <wei@redhat.com>

This patch defines the list of kvm_exit reasons for aarch64. This list is
based on the Exception Class (EC) field of HSR register. With this patch
users can trace the execution of guest VMs better. A sample output from
command "kvm_stat -1 -t" is shown as the following:
<...>
kvm_exit(WATCHPT_HYP)          0         0
kvm_exit(WFI)               9422      9361

NOTE: This patch requires TRACE_EVENT(kvm_exit) to include exit_reason
field in TP_ARGS. A patch to upstream kernel has been submitted.

Signed-off-by: Wei Huang <wei@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 scripts/kvm/kvm_stat | 42 +++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 41 insertions(+), 1 deletion(-)

diff --git a/scripts/kvm/kvm_stat b/scripts/kvm/kvm_stat
index c0c4ff0..c65cabd 100755
--- a/scripts/kvm/kvm_stat
+++ b/scripts/kvm/kvm_stat
@@ -145,6 +145,45 @@ svm_exit_reasons = {
     0x400: 'NPF',
 }
 
+# EC definition of HSR (from arch/arm64/include/asm/kvm_arm.h)
+aarch64_exit_reasons = {
+    0x00: 'UNKNOWN',
+    0x01: 'WFI',
+    0x03: 'CP15_32',
+    0x04: 'CP15_64',
+    0x05: 'CP14_MR',
+    0x06: 'CP14_LS',
+    0x07: 'FP_ASIMD',
+    0x08: 'CP10_ID',
+    0x0C: 'CP14_64',
+    0x0E: 'ILL_ISS',
+    0x11: 'SVC32',
+    0x12: 'HVC32',
+    0x13: 'SMC32',
+    0x15: 'SVC64',
+    0x16: 'HVC64',
+    0x17: 'SMC64',
+    0x18: 'SYS64',
+    0x20: 'IABT',
+    0x21: 'IABT_HYP',
+    0x22: 'PC_ALIGN',
+    0x24: 'DABT',
+    0x25: 'DABT_HYP',
+    0x26: 'SP_ALIGN',
+    0x28: 'FP_EXC32',
+    0x2C: 'FP_EXC64',
+    0x2F: 'SERROR',
+    0x30: 'BREAKPT',
+    0x31: 'BREAKPT_HYP',
+    0x32: 'SOFTSTP',
+    0x33: 'SOFTSTP_HYP',
+    0x34: 'WATCHPT',
+    0x35: 'WATCHPT_HYP',
+    0x38: 'BKPT32',
+    0x3A: 'VECTOR32',
+    0x3C: 'BRK64',
+}
+
 # From include/uapi/linux/kvm.h, KVM_EXIT_xxx
 userspace_exit_reasons = {
      0: 'UNKNOWN',
@@ -212,7 +251,8 @@ def ppc_init():
 
 def aarch64_init():
     globals().update({
-        'sc_perf_evt_open' : 241
+        'sc_perf_evt_open' : 241,
+        'exit_reasons' : aarch64_exit_reasons,
     })
 
 def detect_platform():
-- 
2.3.0

[Qemu-devel] [PULL 02/21] qdev: support to get a device firmware path directly

[Paolo Bonzini]

From: Gonglei <arei.gonglei@huawei.com>

commit 6b1566c (qdev: Introduce FWPathProvider interface) did a
good job for supproting to get firmware path on some different
architectures.

Moreover further more, we can use the interface to get firmware
path name for a device which isn't attached a specific bus,
such as virtio-bus, scsi-bus etc.

When the device (such as vhost-scsi) realize the TYPE_FW_PATH_PROVIDER
interface, we should introduce a new function to get the correct firmware
path name for it.

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 bootdevice.c           | 31 +++++++++++++++++--------------
 hw/core/qdev.c         |  7 +++++++
 include/hw/qdev-core.h |  1 +
 3 files changed, 25 insertions(+), 14 deletions(-)

diff --git a/bootdevice.c b/bootdevice.c
index 5914417..c3a010c 100644
--- a/bootdevice.c
+++ b/bootdevice.c
@@ -210,7 +210,9 @@ char *get_boot_devices_list(size_t *size, bool ignore_suffixes)
     char *list = NULL;
 
     QTAILQ_FOREACH(i, &fw_boot_order, link) {
-        char *devpath = NULL, *bootpath;
+        char *devpath = NULL,  *suffix = NULL;
+        char *bootpath;
+        char *d;
         size_t len;
 
         if (i->dev) {
@@ -218,21 +220,22 @@ char *get_boot_devices_list(size_t *size, bool ignore_suffixes)
             assert(devpath);
         }
 
-        if (i->suffix && !ignore_suffixes && devpath) {
-            size_t bootpathlen = strlen(devpath) + strlen(i->suffix) + 1;
-
-            bootpath = g_malloc(bootpathlen);
-            snprintf(bootpath, bootpathlen, "%s%s", devpath, i->suffix);
-            g_free(devpath);
-        } else if (devpath) {
-            bootpath = devpath;
-        } else if (!ignore_suffixes) {
-            assert(i->suffix);
-            bootpath = g_strdup(i->suffix);
-        } else {
-            bootpath = g_strdup("");
+        if (!ignore_suffixes) {
+            d = qdev_get_own_fw_dev_path_from_handler(i->dev->parent_bus, i->dev);
+            if (d) {
+                assert(!i->suffix);
+                suffix = d;
+            } else {
+                suffix = g_strdup(i->suffix);
+            }
         }
 
+        bootpath = g_strdup_printf("%s%s",
+                                   devpath ? devpath : "",
+                                   suffix ? suffix : "");
+        g_free(devpath);
+        g_free(suffix);
+
         if (total) {
             list[total-1] = '\n';
         }
diff --git a/hw/core/qdev.c b/hw/core/qdev.c
index 2eacac0..44c6b93 100644
--- a/hw/core/qdev.c
+++ b/hw/core/qdev.c
@@ -818,6 +818,13 @@ static char *qdev_get_fw_dev_path_from_handler(BusState *bus, DeviceState *dev)
     return d;
 }
 
+char *qdev_get_own_fw_dev_path_from_handler(BusState *bus, DeviceState *dev)
+{
+    Object *obj = OBJECT(dev);
+
+    return fw_path_provider_try_get_dev_path(obj, bus, dev);
+}
+
 static int qdev_get_fw_dev_path_helper(DeviceState *dev, char *p, int size)
 {
     int l = 0;
diff --git a/include/hw/qdev-core.h b/include/hw/qdev-core.h
index 15a226f..4e673f9 100644
--- a/include/hw/qdev-core.h
+++ b/include/hw/qdev-core.h
@@ -342,6 +342,7 @@ void qbus_reset_all_fn(void *opaque);
 BusState *sysbus_get_default(void);
 
 char *qdev_get_fw_dev_path(DeviceState *dev);
+char *qdev_get_own_fw_dev_path_from_handler(BusState *bus, DeviceState *dev);
 
 /**
  * @qdev_machine_init
-- 
2.3.0

[Qemu-devel] [PULL 03/21] vhost-scsi: add bootindex property

[Paolo Bonzini]

From: Gonglei <arei.gonglei@huawei.com>

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/vhost-scsi.c           | 9 +++++++++
 hw/virtio/virtio-pci.c         | 2 ++
 include/hw/virtio/vhost-scsi.h | 1 +
 3 files changed, 12 insertions(+)

diff --git a/hw/scsi/vhost-scsi.c b/hw/scsi/vhost-scsi.c
index dcb2bc5..9c4f613 100644
--- a/hw/scsi/vhost-scsi.c
+++ b/hw/scsi/vhost-scsi.c
@@ -290,11 +290,20 @@ static void vhost_scsi_class_init(ObjectClass *klass, void *data)
     vdc->set_status = vhost_scsi_set_status;
 }
 
+static void vhost_scsi_instance_init(Object *obj)
+{
+    VHostSCSI *dev = VHOST_SCSI(obj);
+
+    device_add_bootindex_property(obj, &dev->bootindex, "bootindex", NULL,
+                                  DEVICE(dev), NULL);
+}
+
 static const TypeInfo vhost_scsi_info = {
     .name = TYPE_VHOST_SCSI,
     .parent = TYPE_VIRTIO_SCSI_COMMON,
     .instance_size = sizeof(VHostSCSI),
     .class_init = vhost_scsi_class_init,
+    .instance_init = vhost_scsi_instance_init,
 };
 
 static void virtio_register_types(void)
diff --git a/hw/virtio/virtio-pci.c b/hw/virtio/virtio-pci.c
index dde1d73..604cb5b 100644
--- a/hw/virtio/virtio-pci.c
+++ b/hw/virtio/virtio-pci.c
@@ -1238,6 +1238,8 @@ static void vhost_scsi_pci_instance_init(Object *obj)
 
     virtio_instance_init_common(obj, &dev->vdev, sizeof(dev->vdev),
                                 TYPE_VHOST_SCSI);
+    object_property_add_alias(obj, "bootindex", OBJECT(&dev->vdev),
+                              "bootindex", &error_abort);
 }
 
 static const TypeInfo vhost_scsi_pci_info = {
diff --git a/include/hw/virtio/vhost-scsi.h b/include/hw/virtio/vhost-scsi.h
index 85cc031..ed50289 100644
--- a/include/hw/virtio/vhost-scsi.h
+++ b/include/hw/virtio/vhost-scsi.h
@@ -60,6 +60,7 @@ typedef struct VHostSCSI {
     Error *migration_blocker;
 
     struct vhost_dev dev;
+    int32_t bootindex;
 } VHostSCSI;
 
 #define DEFINE_VHOST_SCSI_PROPERTIES(_state, _conf_field) \
-- 
2.3.0

[Qemu-devel] [PULL 04/21] vhost-scsi: expose the TYPE_FW_PATH_PROVIDER interface

[Paolo Bonzini]

From: Gonglei <arei.gonglei@huawei.com>

In the way, we can make the bootindex property take effect.
At the meanwhile, the firmware path name of vhost-scsi is
"channel@channel/vhost-scsi@target,lun".

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/vhost-scsi.c           | 20 ++++++++++++++++++++
 include/hw/virtio/vhost-scsi.h |  3 +++
 2 files changed, 23 insertions(+)

diff --git a/hw/scsi/vhost-scsi.c b/hw/scsi/vhost-scsi.c
index 9c4f613..dc9076e 100644
--- a/hw/scsi/vhost-scsi.c
+++ b/hw/scsi/vhost-scsi.c
@@ -24,6 +24,7 @@
 #include "hw/virtio/virtio-scsi.h"
 #include "hw/virtio/virtio-bus.h"
 #include "hw/virtio/virtio-access.h"
+#include "hw/fw-path-provider.h"
 
 /* Features supported by host kernel. */
 static const int kernel_feature_bits[] = {
@@ -271,6 +272,19 @@ static void vhost_scsi_unrealize(DeviceState *dev, Error **errp)
     virtio_scsi_common_unrealize(dev, errp);
 }
 
+/*
+ * Implementation of an interface to adjust firmware path
+ * for the bootindex property handling.
+ */
+static char *vhost_scsi_get_fw_dev_path(FWPathProvider *p, BusState *bus,
+                                        DeviceState *dev)
+{
+    VHostSCSI *s = VHOST_SCSI(dev);
+    /* format: channel@channel/vhost-scsi@target,lun */
+    return g_strdup_printf("channel@%x/%s@%x,%x", s->channel,
+                           qdev_fw_name(dev), s->target, s->lun);
+}
+
 static Property vhost_scsi_properties[] = {
     DEFINE_VHOST_SCSI_PROPERTIES(VHostSCSI, parent_obj.conf),
     DEFINE_PROP_END_OF_LIST(),
@@ -280,6 +294,7 @@ static void vhost_scsi_class_init(ObjectClass *klass, void *data)
 {
     DeviceClass *dc = DEVICE_CLASS(klass);
     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
+    FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(klass);
 
     dc->props = vhost_scsi_properties;
     set_bit(DEVICE_CATEGORY_STORAGE, dc->categories);
@@ -288,6 +303,7 @@ static void vhost_scsi_class_init(ObjectClass *klass, void *data)
     vdc->get_features = vhost_scsi_get_features;
     vdc->set_config = vhost_scsi_set_config;
     vdc->set_status = vhost_scsi_set_status;
+    fwc->get_dev_path = vhost_scsi_get_fw_dev_path;
 }
 
 static void vhost_scsi_instance_init(Object *obj)
@@ -304,6 +320,10 @@ static const TypeInfo vhost_scsi_info = {
     .instance_size = sizeof(VHostSCSI),
     .class_init = vhost_scsi_class_init,
     .instance_init = vhost_scsi_instance_init,
+    .interfaces = (InterfaceInfo[]) {
+        { TYPE_FW_PATH_PROVIDER },
+        { }
+    },
 };
 
 static void virtio_register_types(void)
diff --git a/include/hw/virtio/vhost-scsi.h b/include/hw/virtio/vhost-scsi.h
index ed50289..c0056c2 100644
--- a/include/hw/virtio/vhost-scsi.h
+++ b/include/hw/virtio/vhost-scsi.h
@@ -61,6 +61,9 @@ typedef struct VHostSCSI {
 
     struct vhost_dev dev;
     int32_t bootindex;
+    int channel;
+    int target;
+    int lun;
 } VHostSCSI;
 
 #define DEFINE_VHOST_SCSI_PROPERTIES(_state, _conf_field) \
-- 
2.3.0

[Qemu-devel] [PULL 05/21] vhost-scsi: add a property for booting

[Paolo Bonzini]

From: Gonglei <arei.gonglei@huawei.com>

Because Qemu only accept an wwpn argument for vhost-scsi, we
cannot assign a tpgt. That's say tpg is transparent for Qemu, Qemu
doesn't know which tpg can boot, but vhost-scsi driver module
doesn't know too for one assigned wwpn.

At present, we assume that the first tpg can boot only, and add
a boot_tpgt property that defaults to 0. Of course, people can
pass a valid value by qemu command line.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 include/hw/virtio/vhost-scsi.h  | 1 +
 include/hw/virtio/virtio-scsi.h | 1 +
 2 files changed, 2 insertions(+)

diff --git a/include/hw/virtio/vhost-scsi.h b/include/hw/virtio/vhost-scsi.h
index c0056c2..dea0075 100644
--- a/include/hw/virtio/vhost-scsi.h
+++ b/include/hw/virtio/vhost-scsi.h
@@ -69,6 +69,7 @@ typedef struct VHostSCSI {
 #define DEFINE_VHOST_SCSI_PROPERTIES(_state, _conf_field) \
     DEFINE_PROP_STRING("vhostfd", _state, _conf_field.vhostfd), \
     DEFINE_PROP_STRING("wwpn", _state, _conf_field.wwpn), \
+    DEFINE_PROP_UINT32("boot_tpgt", _state, _conf_field.boot_tpgt, 0), \
     DEFINE_PROP_UINT32("num_queues", _state, _conf_field.num_queues, 1), \
     DEFINE_PROP_UINT32("max_sectors", _state, _conf_field.max_sectors, 0xFFFF), \
     DEFINE_PROP_UINT32("cmd_per_lun", _state, _conf_field.cmd_per_lun, 128)
diff --git a/include/hw/virtio/virtio-scsi.h b/include/hw/virtio/virtio-scsi.h
index bf17cc9..c122e7a 100644
--- a/include/hw/virtio/virtio-scsi.h
+++ b/include/hw/virtio/virtio-scsi.h
@@ -153,6 +153,7 @@ struct VirtIOSCSIConf {
     uint32_t cmd_per_lun;
     char *vhostfd;
     char *wwpn;
+    uint32_t boot_tpgt;
     IOThread *iothread;
 };
 
-- 
2.3.0

[Qemu-devel] [PULL 06/21] vhost-scsi: set the bootable value of channel/target/lun

[Paolo Bonzini]

From: Gonglei <arei.gonglei@huawei.com>

At present, the target is valued boot_tpgt, In addition,
channel and lun both are 0 for bootable vhost-scsi device.

Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Bo Su <subo7@huawei.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/scsi/vhost-scsi.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/hw/scsi/vhost-scsi.c b/hw/scsi/vhost-scsi.c
index dc9076e..e30ff84 100644
--- a/hw/scsi/vhost-scsi.c
+++ b/hw/scsi/vhost-scsi.c
@@ -251,6 +251,12 @@ static void vhost_scsi_realize(DeviceState *dev, Error **errp)
         return;
     }
 
+    /* At present, channel and lun both are 0 for bootable vhost-scsi disk */
+    s->channel = 0;
+    s->lun = 0;
+    /* Note: we can also get the minimum tpgt from kernel */
+    s->target = vs->conf.boot_tpgt;
+
     error_setg(&s->migration_blocker,
             "vhost-scsi does not support migration");
     migrate_add_blocker(s->migration_blocker);
-- 
2.3.0

[Qemu-devel] [PULL 07/21] rcu: do not let RCU callbacks pile up indefinitely

[Paolo Bonzini]

Always process them within a short time.  Even though waiting a little
is useful, it is not okay to delay e.g. qemu_opts_del forever.

Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Tested-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 util/rcu.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/util/rcu.c b/util/rcu.c
index c9c3e6e..486d7b6 100644
--- a/util/rcu.c
+++ b/util/rcu.c
@@ -223,14 +223,16 @@ static void *call_rcu_thread(void *opaque)
          * Fetch rcu_call_count now, we only must process elements that were
          * added before synchronize_rcu() starts.
          */
-        while (n < RCU_CALL_MIN_SIZE && ++tries <= 5) {
-            g_usleep(100000);
-            qemu_event_reset(&rcu_call_ready_event);
-            n = atomic_read(&rcu_call_count);
-            if (n < RCU_CALL_MIN_SIZE) {
-                qemu_event_wait(&rcu_call_ready_event);
+        while (n == 0 || (n < RCU_CALL_MIN_SIZE && ++tries <= 5)) {
+            g_usleep(10000);
+            if (n == 0) {
+                qemu_event_reset(&rcu_call_ready_event);
                 n = atomic_read(&rcu_call_count);
+                if (n == 0) {
+                    qemu_event_wait(&rcu_call_ready_event);
+                }
             }
+            n = atomic_read(&rcu_call_count);
         }
 
         atomic_sub(&rcu_call_count, n);
-- 
2.3.0

[Qemu-devel] [PULL 08/21] rcu: run RCU callbacks under the BQL

[Paolo Bonzini]

This needs to go away sooner or later, but one complication is the
complex VFIO data structures that are modified in instance_finalize.
Take a shortcut for now.

Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Tested-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 tests/Makefile | 2 +-
 util/rcu.c     | 5 +++++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/tests/Makefile b/tests/Makefile
index 5caccf7..06acfa1 100644
--- a/tests/Makefile
+++ b/tests/Makefile
@@ -255,7 +255,7 @@ tests/test-x86-cpuid$(EXESUF): tests/test-x86-cpuid.o
 tests/test-xbzrle$(EXESUF): tests/test-xbzrle.o migration/xbzrle.o page_cache.o libqemuutil.a
 tests/test-cutils$(EXESUF): tests/test-cutils.o util/cutils.o
 tests/test-int128$(EXESUF): tests/test-int128.o
-tests/rcutorture$(EXESUF): tests/rcutorture.o libqemuutil.a
+tests/rcutorture$(EXESUF): tests/rcutorture.o libqemuutil.a libqemustub.a
 
 tests/test-qdev-global-props$(EXESUF): tests/test-qdev-global-props.o \
 	hw/core/qdev.o hw/core/qdev-properties.o hw/core/hotplug.o\
diff --git a/util/rcu.c b/util/rcu.c
index 486d7b6..bd73b8e 100644
--- a/util/rcu.c
+++ b/util/rcu.c
@@ -35,6 +35,7 @@
 #include "qemu/rcu.h"
 #include "qemu/atomic.h"
 #include "qemu/thread.h"
+#include "qemu/main-loop.h"
 
 /*
  * Global grace period counter.  Bit 0 is always one in rcu_gp_ctr.
@@ -237,20 +238,24 @@ static void *call_rcu_thread(void *opaque)
 
         atomic_sub(&rcu_call_count, n);
         synchronize_rcu();
+        qemu_mutex_lock_iothread();
         while (n > 0) {
             node = try_dequeue();
             while (!node) {
+                qemu_mutex_unlock_iothread();
                 qemu_event_reset(&rcu_call_ready_event);
                 node = try_dequeue();
                 if (!node) {
                     qemu_event_wait(&rcu_call_ready_event);
                     node = try_dequeue();
                 }
+                qemu_mutex_lock_iothread();
             }
 
             n--;
             node->func(node);
         }
+        qemu_mutex_unlock_iothread();
     }
     abort();
 }
-- 
2.3.0

[Qemu-devel] [PULL 09/21] memory: keep the owner of the AddressSpace alive until do_address_space_destroy

[Paolo Bonzini]

This fixes a use-after-free if do_address_space_destroy is executed
too late.

Reviewed-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Tested-by: Michael Roth <mdroth@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 memory.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/memory.c b/memory.c
index 130152c..20f6d9e 100644
--- a/memory.c
+++ b/memory.c
@@ -1943,6 +1943,7 @@ void memory_listener_unregister(MemoryListener *listener)
 
 void address_space_init(AddressSpace *as, MemoryRegion *root, const char *name)
 {
+    memory_region_ref(root);
     memory_region_transaction_begin();
     as->root = root;
     as->current_map = g_new(FlatView, 1);
@@ -1969,10 +1970,13 @@ static void do_address_space_destroy(AddressSpace *as)
     flatview_unref(as->current_map);
     g_free(as->name);
     g_free(as->ioeventfds);
+    memory_region_unref(as->root);
 }
 
 void address_space_destroy(AddressSpace *as)
 {
+    MemoryRegion *root = as->root;
+
     /* Flush out anything from MemoryListeners listening in on this */
     memory_region_transaction_begin();
     as->root = NULL;
@@ -1984,6 +1988,7 @@ void address_space_destroy(AddressSpace *as)
      * entries that the guest should never use.  Wait for the old
      * values to expire before freeing the data.
      */
+    as->root = root;
     call_rcu(as, do_address_space_destroy, rcu);
 }
 
-- 
2.3.0

[Qemu-devel] [PULL 10/21] pcie: remove mmconfig memory leak and wrap mmconfig update with transaction

[Paolo Bonzini]

This memory leak was introduced inadvertently by omitting object_unparent.
A better fix is to use the new memory_region_set_size instead of destroying
and recreating the MMIO region on the fly.

Also, ensure that unmapping and remapping the region is done atomically.

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Igor Mammedov <imammedo@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/pci/pcie_host.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/hw/pci/pcie_host.c b/hw/pci/pcie_host.c
index dfb4a2b..d8afba8 100644
--- a/hw/pci/pcie_host.c
+++ b/hw/pci/pcie_host.c
@@ -88,6 +88,8 @@ static void pcie_host_init(Object *obj)
     PCIExpressHost *e = PCIE_HOST_BRIDGE(obj);
 
     e->base_addr = PCIE_BASE_ADDR_UNMAPPED;
+    memory_region_init_io(&e->mmio, OBJECT(e), &pcie_mmcfg_ops, e, "pcie-mmcfg-mmio",
+                          PCIE_MMCFG_SIZE_MAX);
 }
 
 void pcie_host_mmcfg_unmap(PCIExpressHost *e)
@@ -104,8 +106,7 @@ void pcie_host_mmcfg_init(PCIExpressHost *e, uint32_t size)
     assert(size >= PCIE_MMCFG_SIZE_MIN);
     assert(size <= PCIE_MMCFG_SIZE_MAX);
     e->size = size;
-    memory_region_init_io(&e->mmio, OBJECT(e), &pcie_mmcfg_ops, e,
-                          "pcie-mmcfg", e->size);
+    memory_region_set_size(&e->mmio, e->size);
 }
 
 void pcie_host_mmcfg_map(PCIExpressHost *e, hwaddr addr,
@@ -121,10 +122,12 @@ void pcie_host_mmcfg_update(PCIExpressHost *e,
                             hwaddr addr,
                             uint32_t size)
 {
+    memory_region_transaction_begin();
     pcie_host_mmcfg_unmap(e);
     if (enable) {
         pcie_host_mmcfg_map(e, addr, size);
     }
+    memory_region_transaction_commit();
 }
 
 static const TypeInfo pcie_host_type_info = {
-- 
2.3.0

[Qemu-devel] [PULL 11/21] pci: split shpc_cleanup and shpc_free

[Paolo Bonzini]

object_unparent should not be called until the parent device is going to be
destroyed.  Only remove the capability and do memory_region_del_subregion
at unrealize time.  Freeing the data structures is left in shpc_free, to
be called from the instance_finalize callback.

Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Matthew Rosato <mjrosato@linux.vnet.ibm.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/pci-bridge/pci_bridge_dev.c | 14 ++++++++++----
 hw/pci/shpc.c                  | 11 ++++++++++-
 include/hw/pci/shpc.h          |  1 +
 3 files changed, 21 insertions(+), 5 deletions(-)

diff --git a/hw/pci-bridge/pci_bridge_dev.c b/hw/pci-bridge/pci_bridge_dev.c
index 252ea5e..36f73e1 100644
--- a/hw/pci-bridge/pci_bridge_dev.c
+++ b/hw/pci-bridge/pci_bridge_dev.c
@@ -97,6 +97,11 @@ static void pci_bridge_dev_exitfn(PCIDevice *dev)
     pci_bridge_exitfn(dev);
 }
 
+static void pci_bridge_dev_instance_finalize(Object *obj)
+{
+    shpc_free(PCI_DEVICE(obj));
+}
+
 static void pci_bridge_dev_write_config(PCIDevice *d,
                                         uint32_t address, uint32_t val, int len)
 {
@@ -154,10 +159,11 @@ static void pci_bridge_dev_class_init(ObjectClass *klass, void *data)
 }
 
 static const TypeInfo pci_bridge_dev_info = {
-    .name          = TYPE_PCI_BRIDGE_DEV,
-    .parent        = TYPE_PCI_BRIDGE,
-    .instance_size = sizeof(PCIBridgeDev),
-    .class_init = pci_bridge_dev_class_init,
+    .name              = TYPE_PCI_BRIDGE_DEV,
+    .parent            = TYPE_PCI_BRIDGE,
+    .instance_size     = sizeof(PCIBridgeDev),
+    .class_init        = pci_bridge_dev_class_init,
+    .instance_finalize = pci_bridge_dev_instance_finalize,
     .interfaces = (InterfaceInfo[]) {
         { TYPE_HOTPLUG_HANDLER },
         { }
diff --git a/hw/pci/shpc.c b/hw/pci/shpc.c
index 27c496e..5fd7f4b 100644
--- a/hw/pci/shpc.c
+++ b/hw/pci/shpc.c
@@ -663,13 +663,22 @@ void shpc_cleanup(PCIDevice *d, MemoryRegion *bar)
     SHPCDevice *shpc = d->shpc;
     d->cap_present &= ~QEMU_PCI_CAP_SHPC;
     memory_region_del_subregion(bar, &shpc->mmio);
-    object_unparent(OBJECT(&shpc->mmio));
     /* TODO: cleanup config space changes? */
+}
+
+void shpc_free(PCIDevice *d)
+{
+    SHPCDevice *shpc = d->shpc;
+    if (!shpc) {
+        return;
+    }
+    object_unparent(OBJECT(&shpc->mmio));
     g_free(shpc->config);
     g_free(shpc->cmask);
     g_free(shpc->wmask);
     g_free(shpc->w1cmask);
     g_free(shpc);
+    d->shpc = NULL;
 }
 
 void shpc_cap_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int l)
diff --git a/include/hw/pci/shpc.h b/include/hw/pci/shpc.h
index 025bc5b..9bbea39 100644
--- a/include/hw/pci/shpc.h
+++ b/include/hw/pci/shpc.h
@@ -41,6 +41,7 @@ void shpc_reset(PCIDevice *d);
 int shpc_bar_size(PCIDevice *dev);
 int shpc_init(PCIDevice *dev, PCIBus *sec_bus, MemoryRegion *bar, unsigned off);
 void shpc_cleanup(PCIDevice *dev, MemoryRegion *bar);
+void shpc_free(PCIDevice *dev);
 void shpc_cap_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int len);
 
 
-- 
2.3.0

[Qemu-devel] [PULL 12/21] docs: clarify memory region lifecycle

[Paolo Bonzini]

Now that objects actually obey the rules, document them.

Reviewed-by: Matthew Rosato <mjrosato@linux.vnet.ibm.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 docs/memory.txt | 74 ++++++++++++++++++++++++++++++++++++++++++++-------------
 1 file changed, 58 insertions(+), 16 deletions(-)

diff --git a/docs/memory.txt b/docs/memory.txt
index b12f1f0..2ceb348 100644
--- a/docs/memory.txt
+++ b/docs/memory.txt
@@ -73,17 +73,66 @@ stability.
 Region lifecycle
 ----------------
 
-A region is created by one of the constructor functions (memory_region_init*())
-and attached to an object.  It is then destroyed by object_unparent() or simply
-when the parent object dies.
+A region is created by one of the memory_region_init*() functions and
+attached to an object, which acts as its owner or parent.  QEMU ensures
+that the owner object remains alive as long as the region is visible to
+the guest, or as long as the region is in use by a virtual CPU or another
+device.  For example, the owner object will not die between an
+address_space_map operation and the corresponding address_space_unmap.
 
-In between, a region can be added to an address space
-by using memory_region_add_subregion() and removed using
-memory_region_del_subregion().  Destroying the region implicitly
-removes the region from the address space.
+After creation, a region can be added to an address space or a
+container with memory_region_add_subregion(), and removed using
+memory_region_del_subregion().
+
+Various region attributes (read-only, dirty logging, coalesced mmio,
+ioeventfd) can be changed during the region lifecycle.  They take effect
+as soon as the region is made visible.  This can be immediately, later,
+or never.
+
+Destruction of a memory region happens automatically when the owner
+object dies.
+
+If however the memory region is part of a dynamically allocated data
+structure, you should call object_unparent() to destroy the memory region
+before the data structure is freed.  For an example see VFIOMSIXInfo
+and VFIOQuirk in hw/vfio/pci.c.
+
+You must not destroy a memory region as long as it may be in use by a
+device or CPU.  In order to do this, as a general rule do not create or
+destroy memory regions dynamically during a device's lifetime, and only
+call object_unparent() in the memory region owner's instance_finalize
+callback.  The dynamically allocated data structure that contains the
+memory region then should obviously be freed in the instance_finalize
+callback as well.
+
+If you break this rule, the following situation can happen:
+
+- the memory region's owner had a reference taken via memory_region_ref
+  (for example by address_space_map)
+
+- the region is unparented, and has no owner anymore
+
+- when address_space_unmap is called, the reference to the memory region's
+  owner is leaked.
+
+
+There is an exception to the above rule: it is okay to call
+object_unparent at any time for an alias or a container region.  It is
+therefore also okay to create or destroy alias and container regions
+dynamically during a device's lifetime.
+
+This exceptional usage is valid because aliases and containers only help
+QEMU building the guest's memory map; they are never accessed directly.
+memory_region_ref and memory_region_unref are never called on aliases
+or containers, and the above situation then cannot happen.  Exploiting
+this exception is rarely necessary, and therefore it is discouraged,
+but nevertheless it is used in a few places.
+
+For regions that "have no owner" (NULL is passed at creation time), the
+machine object is actually used as the owner.  Since instance_finalize is
+never called for the machine object, you must never call object_unparent
+on regions that have no owner, unless they are aliases or containers.
 
-Region attributes may be changed at any point; they take effect once
-the region becomes exposed to the guest.
 
 Overlapping regions and priority
 --------------------------------
@@ -215,13 +264,6 @@ BAR containing MMIO registers is mapped after it.
 Note that if the guest maps a BAR outside the PCI hole, it would not be
 visible as the pci-hole alias clips it to a 0.5GB range.
 
-Attributes
-----------
-
-Various region attributes (read-only, dirty logging, coalesced mmio, ioeventfd)
-can be changed during the region lifecycle.  They take effect once the region
-is made visible (which can be immediately, later, or never).
-
 MMIO Operations
 ---------------
 
-- 
2.3.0

[Qemu-devel] [PULL 13/21] exec: introduce cpu_reload_memory_map

[Paolo Bonzini]

This for now is a simple TLB flush.  This can change later for two
reasons:

1) an AddressSpaceDispatch will be cached in the CPUState object

2) it will not be possible to do tlb_flush once the TCG-generated code
runs outside the BQL.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 cpu-exec.c              | 6 ++++++
 exec.c                  | 2 +-
 include/exec/exec-all.h | 1 +
 3 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/cpu-exec.c b/cpu-exec.c
index fa506e6..78fe382 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -141,6 +141,12 @@ void cpu_resume_from_signal(CPUState *cpu, void *puc)
     cpu->exception_index = -1;
     siglongjmp(cpu->jmp_env, 1);
 }
+
+void cpu_reload_memory_map(CPUState *cpu)
+{
+    /* The TLB is protected by the iothread lock.  */
+    tlb_flush(cpu, 1);
+}
 #endif
 
 /* Execute a TB, and fix up the CPU state afterwards if necessary */
diff --git a/exec.c b/exec.c
index 6dff7bc..2bfb4d3 100644
--- a/exec.c
+++ b/exec.c
@@ -2026,7 +2026,7 @@ static void tcg_commit(MemoryListener *listener)
         if (cpu->tcg_as_listener != listener) {
             continue;
         }
-        tlb_flush(cpu, 1);
+        cpu_reload_memory_map(cpu);
     }
 }
 
diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index 6a15448..1b30813 100644
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -96,6 +96,7 @@ void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
 void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end,
                               int is_cpu_write_access);
 #if !defined(CONFIG_USER_ONLY)
+void cpu_reload_memory_map(CPUState *cpu);
 void tcg_cpu_address_space_init(CPUState *cpu, AddressSpace *as);
 /* cputlb.c */
 void tlb_flush_page(CPUState *cpu, target_ulong addr);
-- 
2.3.0

[Qemu-devel] [PULL 14/21] exec: make iotlb RCU-friendly

[Paolo Bonzini]

After the previous patch, TLBs will be flushed on every change to
the memory mapping.  This patch augments that with synchronization
of the MemoryRegionSections referred to in the iotlb array.

With this change, it is guaranteed that iotlb_to_region will access
the correct memory map, even once the TLB will be accessed outside
the BQL.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 cpu-exec.c              |  6 +++++-
 cputlb.c                |  5 ++---
 exec.c                  | 13 ++++++++-----
 include/exec/cputlb.h   |  2 +-
 include/exec/exec-all.h |  3 ++-
 include/qom/cpu.h       |  1 +
 softmmu_template.h      |  4 ++--
 7 files changed, 21 insertions(+), 13 deletions(-)

diff --git a/cpu-exec.c b/cpu-exec.c
index 78fe382..98f968d 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -24,6 +24,8 @@
 #include "qemu/atomic.h"
 #include "sysemu/qtest.h"
 #include "qemu/timer.h"
+#include "exec/address-spaces.h"
+#include "exec/memory-internal.h"
 
 /* -icount align implementation. */
 
@@ -144,7 +146,9 @@ void cpu_resume_from_signal(CPUState *cpu, void *puc)
 
 void cpu_reload_memory_map(CPUState *cpu)
 {
-    /* The TLB is protected by the iothread lock.  */
+    /* The CPU and TLB are protected by the iothread lock.  */
+    AddressSpaceDispatch *d = cpu->as->dispatch;
+    cpu->memory_dispatch = d;
     tlb_flush(cpu, 1);
 }
 #endif
diff --git a/cputlb.c b/cputlb.c
index 3b271d4..f92db5e 100644
--- a/cputlb.c
+++ b/cputlb.c
@@ -265,8 +265,7 @@ void tlb_set_page(CPUState *cpu, target_ulong vaddr,
     }
 
     sz = size;
-    section = address_space_translate_for_iotlb(cpu->as, paddr,
-                                                &xlat, &sz);
+    section = address_space_translate_for_iotlb(cpu, paddr, &xlat, &sz);
     assert(sz >= TARGET_PAGE_SIZE);
 
 #if defined(DEBUG_TLB)
@@ -347,7 +346,7 @@ tb_page_addr_t get_page_addr_code(CPUArchState *env1, target_ulong addr)
         cpu_ldub_code(env1, addr);
     }
     pd = env1->iotlb[mmu_idx][page_index] & ~TARGET_PAGE_MASK;
-    mr = iotlb_to_region(cpu->as, pd);
+    mr = iotlb_to_region(cpu, pd);
     if (memory_region_is_unassigned(mr)) {
         CPUClass *cc = CPU_GET_CLASS(cpu);
 
diff --git a/exec.c b/exec.c
index 2bfb4d3..fe1e60a 100644
--- a/exec.c
+++ b/exec.c
@@ -401,11 +401,12 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
 }
 
 MemoryRegionSection *
-address_space_translate_for_iotlb(AddressSpace *as, hwaddr addr, hwaddr *xlat,
-                                  hwaddr *plen)
+address_space_translate_for_iotlb(CPUState *cpu, hwaddr addr,
+                                  hwaddr *xlat, hwaddr *plen)
 {
     MemoryRegionSection *section;
-    section = address_space_translate_internal(as->dispatch, addr, xlat, plen, false);
+    section = address_space_translate_internal(cpu->memory_dispatch,
+                                               addr, xlat, plen, false);
 
     assert(!section->mr->iommu_ops);
     return section;
@@ -1961,9 +1962,11 @@ static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
     return phys_section_add(map, &section);
 }
 
-MemoryRegion *iotlb_to_region(AddressSpace *as, hwaddr index)
+MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index)
 {
-    return as->dispatch->map.sections[index & ~TARGET_PAGE_MASK].mr;
+    MemoryRegionSection *sections = cpu->memory_dispatch->map.sections;
+
+    return sections[index & ~TARGET_PAGE_MASK].mr;
 }
 
 static void io_mem_init(void)
diff --git a/include/exec/cputlb.h b/include/exec/cputlb.h
index b8ecd6f..e0da9d7 100644
--- a/include/exec/cputlb.h
+++ b/include/exec/cputlb.h
@@ -34,7 +34,7 @@ extern int tlb_flush_count;
 void tb_flush_jmp_cache(CPUState *cpu, target_ulong addr);
 
 MemoryRegionSection *
-address_space_translate_for_iotlb(AddressSpace *as, hwaddr addr, hwaddr *xlat,
+address_space_translate_for_iotlb(CPUState *cpu, hwaddr addr, hwaddr *xlat,
                                   hwaddr *plen);
 hwaddr memory_region_section_get_iotlb(CPUState *cpu,
                                        MemoryRegionSection *section,
diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index 1b30813..bb3fd37 100644
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -338,7 +338,8 @@ extern uintptr_t tci_tb_ptr;
 
 void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align));
 
-struct MemoryRegion *iotlb_to_region(AddressSpace *as, hwaddr index);
+struct MemoryRegion *iotlb_to_region(CPUState *cpu,
+                                     hwaddr index);
 bool io_mem_read(struct MemoryRegion *mr, hwaddr addr,
                  uint64_t *pvalue, unsigned size);
 bool io_mem_write(struct MemoryRegion *mr, hwaddr addr,
diff --git a/include/qom/cpu.h b/include/qom/cpu.h
index 2098f1c..48fd6fb 100644
--- a/include/qom/cpu.h
+++ b/include/qom/cpu.h
@@ -256,6 +256,7 @@ struct CPUState {
     sigjmp_buf jmp_env;
 
     AddressSpace *as;
+    struct AddressSpaceDispatch *memory_dispatch;
     MemoryListener *tcg_as_listener;
 
     void *env_ptr; /* CPUArchState */
diff --git a/softmmu_template.h b/softmmu_template.h
index 6b4e615..0e3dd35 100644
--- a/softmmu_template.h
+++ b/softmmu_template.h
@@ -149,7 +149,7 @@ static inline DATA_TYPE glue(io_read, SUFFIX)(CPUArchState *env,
 {
     uint64_t val;
     CPUState *cpu = ENV_GET_CPU(env);
-    MemoryRegion *mr = iotlb_to_region(cpu->as, physaddr);
+    MemoryRegion *mr = iotlb_to_region(cpu, physaddr);
 
     physaddr = (physaddr & TARGET_PAGE_MASK) + addr;
     cpu->mem_io_pc = retaddr;
@@ -369,7 +369,7 @@ static inline void glue(io_write, SUFFIX)(CPUArchState *env,
                                           uintptr_t retaddr)
 {
     CPUState *cpu = ENV_GET_CPU(env);
-    MemoryRegion *mr = iotlb_to_region(cpu->as, physaddr);
+    MemoryRegion *mr = iotlb_to_region(cpu, physaddr);
 
     physaddr = (physaddr & TARGET_PAGE_MASK) + addr;
     if (mr != &io_mem_rom && mr != &io_mem_notdirty && !cpu_can_do_io(cpu)) {
-- 
2.3.0

[Qemu-devel] [PULL 15/21] exec: RCUify AddressSpaceDispatch

[Paolo Bonzini]

Note that even after this patch, most callers of address_space_*
functions must still be under the big QEMU lock, otherwise the memory
region returned by address_space_translate can disappear as soon as
address_space_translate returns.  This will be fixed in the next part
of this series.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 cpu-exec.c              | 25 ++++++++++++++++++++++++-
 cpus.c                  |  2 +-
 cputlb.c                |  8 ++++++--
 exec.c                  | 34 ++++++++++++++++++++++++++--------
 hw/i386/intel_iommu.c   |  3 +++
 hw/pci-host/apb.c       |  1 +
 hw/ppc/spapr_iommu.c    |  1 +
 include/exec/exec-all.h |  1 +
 8 files changed, 63 insertions(+), 12 deletions(-)

diff --git a/cpu-exec.c b/cpu-exec.c
index 98f968d..adb939a 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -26,6 +26,7 @@
 #include "qemu/timer.h"
 #include "exec/address-spaces.h"
 #include "exec/memory-internal.h"
+#include "qemu/rcu.h"
 
 /* -icount align implementation. */
 
@@ -146,8 +147,27 @@ void cpu_resume_from_signal(CPUState *cpu, void *puc)
 
 void cpu_reload_memory_map(CPUState *cpu)
 {
+    AddressSpaceDispatch *d;
+
+    if (qemu_in_vcpu_thread()) {
+        /* Do not let the guest prolong the critical section as much as it
+         * as it desires.
+         *
+         * Currently, this is prevented by the I/O thread's periodinc kicking
+         * of the VCPU thread (iothread_requesting_mutex, qemu_cpu_kick_thread)
+         * but this will go away once TCG's execution moves out of the global
+         * mutex.
+         *
+         * This pair matches cpu_exec's rcu_read_lock()/rcu_read_unlock(), which
+         * only protects cpu->as->dispatch.  Since we reload it below, we can
+         * split the critical section.
+         */
+        rcu_read_unlock();
+        rcu_read_lock();
+    }
+
     /* The CPU and TLB are protected by the iothread lock.  */
-    AddressSpaceDispatch *d = cpu->as->dispatch;
+    d = atomic_rcu_read(&cpu->as->dispatch);
     cpu->memory_dispatch = d;
     tlb_flush(cpu, 1);
 }
@@ -362,6 +382,8 @@ int cpu_exec(CPUArchState *env)
      * an instruction scheduling constraint on modern architectures.  */
     smp_mb();
 
+    rcu_read_lock();
+
     if (unlikely(exit_request)) {
         cpu->exit_request = 1;
     }
@@ -564,6 +586,7 @@ int cpu_exec(CPUArchState *env)
     } /* for(;;) */
 
     cc->cpu_exec_exit(cpu);
+    rcu_read_unlock();
 
     /* fail safe : never use current_cpu outside cpu_exec() */
     current_cpu = NULL;
diff --git a/cpus.c b/cpus.c
index 0cdd1d7..b826fac 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1104,7 +1104,7 @@ bool qemu_cpu_is_self(CPUState *cpu)
     return qemu_thread_is_self(cpu->thread);
 }
 
-static bool qemu_in_vcpu_thread(void)
+bool qemu_in_vcpu_thread(void)
 {
     return current_cpu && qemu_cpu_is_self(current_cpu);
 }
diff --git a/cputlb.c b/cputlb.c
index f92db5e..38f2151 100644
--- a/cputlb.c
+++ b/cputlb.c
@@ -243,8 +243,12 @@ static void tlb_add_large_page(CPUArchState *env, target_ulong vaddr,
 }
 
 /* Add a new TLB entry. At most one entry for a given virtual address
-   is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
-   supplied size is only used by tlb_flush_page.  */
+ * is permitted. Only a single TARGET_PAGE_SIZE region is mapped, the
+ * supplied size is only used by tlb_flush_page.
+ *
+ * Called from TCG-generated code, which is under an RCU read-side
+ * critical section.
+ */
 void tlb_set_page(CPUState *cpu, target_ulong vaddr,
                   hwaddr paddr, int prot,
                   int mmu_idx, target_ulong size)
diff --git a/exec.c b/exec.c
index fe1e60a..76b3b6c 100644
--- a/exec.c
+++ b/exec.c
@@ -115,6 +115,8 @@ struct PhysPageEntry {
 typedef PhysPageEntry Node[P_L2_SIZE];
 
 typedef struct PhysPageMap {
+    struct rcu_head rcu;
+
     unsigned sections_nb;
     unsigned sections_nb_alloc;
     unsigned nodes_nb;
@@ -124,6 +126,8 @@ typedef struct PhysPageMap {
 } PhysPageMap;
 
 struct AddressSpaceDispatch {
+    struct rcu_head rcu;
+
     /* This is a multi-level map on the physical address space.
      * The bottom level has pointers to MemoryRegionSections.
      */
@@ -315,6 +319,7 @@ bool memory_region_is_unassigned(MemoryRegion *mr)
         && mr != &io_mem_watch;
 }
 
+/* Called from RCU critical section */
 static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
                                                         hwaddr addr,
                                                         bool resolve_subpage)
@@ -330,6 +335,7 @@ static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d,
     return section;
 }
 
+/* Called from RCU critical section */
 static MemoryRegionSection *
 address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat,
                                  hwaddr *plen, bool resolve_subpage)
@@ -370,8 +376,10 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
     MemoryRegion *mr;
     hwaddr len = *plen;
 
+    rcu_read_lock();
     for (;;) {
-        section = address_space_translate_internal(as->dispatch, addr, &addr, plen, true);
+        AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch);
+        section = address_space_translate_internal(d, addr, &addr, plen, true);
         mr = section->mr;
 
         if (!mr->iommu_ops) {
@@ -397,9 +405,11 @@ MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr,
 
     *plen = len;
     *xlat = addr;
+    rcu_read_unlock();
     return mr;
 }
 
+/* Called from RCU critical section */
 MemoryRegionSection *
 address_space_translate_for_iotlb(CPUState *cpu, hwaddr addr,
                                   hwaddr *xlat, hwaddr *plen)
@@ -852,6 +862,7 @@ static void cpu_physical_memory_set_dirty_tracking(bool enable)
     in_migration = enable;
 }
 
+/* Called from RCU critical section */
 hwaddr memory_region_section_get_iotlb(CPUState *cpu,
                                        MemoryRegionSection *section,
                                        target_ulong vaddr,
@@ -1964,7 +1975,8 @@ static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as,
 
 MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index)
 {
-    MemoryRegionSection *sections = cpu->memory_dispatch->map.sections;
+    AddressSpaceDispatch *d = atomic_rcu_read(&cpu->memory_dispatch);
+    MemoryRegionSection *sections = d->map.sections;
 
     return sections[index & ~TARGET_PAGE_MASK].mr;
 }
@@ -2000,6 +2012,12 @@ static void mem_begin(MemoryListener *listener)
     as->next_dispatch = d;
 }
 
+static void address_space_dispatch_free(AddressSpaceDispatch *d)
+{
+    phys_sections_free(&d->map);
+    g_free(d);
+}
+
 static void mem_commit(MemoryListener *listener)
 {
     AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener);
@@ -2008,11 +2026,9 @@ static void mem_commit(MemoryListener *listener)
 
     phys_page_compact_all(next, next->map.nodes_nb);
 
-    as->dispatch = next;
-
+    atomic_rcu_set(&as->dispatch, next);
     if (cur) {
-        phys_sections_free(&cur->map);
-        g_free(cur);
+        call_rcu(cur, address_space_dispatch_free, rcu);
     }
 }
 
@@ -2071,8 +2087,10 @@ void address_space_destroy_dispatch(AddressSpace *as)
 {
     AddressSpaceDispatch *d = as->dispatch;
 
-    g_free(d);
-    as->dispatch = NULL;
+    atomic_rcu_set(&as->dispatch, NULL);
+    if (d) {
+        call_rcu(d, address_space_dispatch_free, rcu);
+    }
 }
 
 static void memory_map_init(void)
diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
index 0a4282a..7da70ff 100644
--- a/hw/i386/intel_iommu.c
+++ b/hw/i386/intel_iommu.c
@@ -745,6 +745,9 @@ static inline bool vtd_is_interrupt_addr(hwaddr addr)
 
 /* Map dev to context-entry then do a paging-structures walk to do a iommu
  * translation.
+ *
+ * Called from RCU critical section.
+ *
  * @bus_num: The bus number
  * @devfn: The devfn, which is the  combined of device and function number
  * @is_write: The access is a write operation
diff --git a/hw/pci-host/apb.c b/hw/pci-host/apb.c
index f573875..832b6c7 100644
--- a/hw/pci-host/apb.c
+++ b/hw/pci-host/apb.c
@@ -205,6 +205,7 @@ static AddressSpace *pbm_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn)
     return &is->iommu_as;
 }
 
+/* Called from RCU critical section */
 static IOMMUTLBEntry pbm_translate_iommu(MemoryRegion *iommu, hwaddr addr,
                                          bool is_write)
 {
diff --git a/hw/ppc/spapr_iommu.c b/hw/ppc/spapr_iommu.c
index da47474..ba003da 100644
--- a/hw/ppc/spapr_iommu.c
+++ b/hw/ppc/spapr_iommu.c
@@ -59,6 +59,7 @@ static sPAPRTCETable *spapr_tce_find_by_liobn(uint32_t liobn)
     return NULL;
 }
 
+/* Called from RCU critical section */
 static IOMMUTLBEntry spapr_tce_translate_iommu(MemoryRegion *iommu, hwaddr addr,
                                                bool is_write)
 {
diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h
index bb3fd37..8eb0db3 100644
--- a/include/exec/exec-all.h
+++ b/include/exec/exec-all.h
@@ -96,6 +96,7 @@ void tb_invalidate_phys_page_range(tb_page_addr_t start, tb_page_addr_t end,
 void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end,
                               int is_cpu_write_access);
 #if !defined(CONFIG_USER_ONLY)
+bool qemu_in_vcpu_thread(void);
 void cpu_reload_memory_map(CPUState *cpu);
 void tcg_cpu_address_space_init(CPUState *cpu, AddressSpace *as);
 /* cputlb.c */
-- 
2.3.0

[Qemu-devel] [PULL 16/21] rcu: introduce RCU-enabled QLIST

[Paolo Bonzini]

From: Mike Day <ncmike@ncultra.org>

Add RCU-enabled variants on the existing bsd DQ facility. Each
operation has the same interface as the existing (non-RCU)
version. Also, each operation is implemented as macro.

Using the RCU-enabled QLIST, existing QLIST users will be able to
convert to RCU without using a different list interface.

Signed-off-by: Mike Day <ncmike@ncultra.org>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 hw/9pfs/virtio-9p-synth.c |   2 +-
 include/qemu/queue.h      |  11 --
 include/qemu/rcu_queue.h  | 134 ++++++++++++++++++++
 tests/Makefile            |   5 +-
 tests/test-rcu-list.c     | 306 ++++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 445 insertions(+), 13 deletions(-)
 create mode 100644 include/qemu/rcu_queue.h
 create mode 100644 tests/test-rcu-list.c

diff --git a/hw/9pfs/virtio-9p-synth.c b/hw/9pfs/virtio-9p-synth.c
index e75aa87..a0ab9a8 100644
--- a/hw/9pfs/virtio-9p-synth.c
+++ b/hw/9pfs/virtio-9p-synth.c
@@ -18,7 +18,7 @@
 #include "fsdev/qemu-fsdev.h"
 #include "virtio-9p-synth.h"
 #include "qemu/rcu.h"
-
+#include "qemu/rcu_queue.h"
 #include <sys/stat.h>
 
 /* Root node for synth file system */
diff --git a/include/qemu/queue.h b/include/qemu/queue.h
index c602797..8094150 100644
--- a/include/qemu/queue.h
+++ b/include/qemu/queue.h
@@ -139,17 +139,6 @@ struct {                                                                \
         (elm)->field.le_prev = &(head)->lh_first;                       \
 } while (/*CONSTCOND*/0)
 
-#define QLIST_INSERT_HEAD_RCU(head, elm, field) do {                    \
-        (elm)->field.le_prev = &(head)->lh_first;                       \
-        (elm)->field.le_next = (head)->lh_first;                        \
-        smp_wmb(); /* fill elm before linking it */                     \
-        if ((head)->lh_first != NULL)  {                                \
-            (head)->lh_first->field.le_prev = &(elm)->field.le_next;    \
-        }                                                               \
-        (head)->lh_first = (elm);                                       \
-        smp_wmb();                                                      \
-} while (/* CONSTCOND*/0)
-
 #define QLIST_REMOVE(elm, field) do {                                   \
         if ((elm)->field.le_next != NULL)                               \
                 (elm)->field.le_next->field.le_prev =                   \
diff --git a/include/qemu/rcu_queue.h b/include/qemu/rcu_queue.h
new file mode 100644
index 0000000..3aca7a5
--- /dev/null
+++ b/include/qemu/rcu_queue.h
@@ -0,0 +1,134 @@
+#ifndef QEMU_RCU_QUEUE_H
+#define QEMU_RCU_QUEUE_H
+
+/*
+ * rcu_queue.h
+ *
+ * RCU-friendly versions of the queue.h primitives.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ *
+ * Copyright (c) 2013 Mike D. Day, IBM Corporation.
+ *
+ * IBM's contributions to this file may be relicensed under LGPLv2 or later.
+ */
+
+#include "qemu/queue.h"
+#include "qemu/atomic.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/*
+ * List access methods.
+ */
+#define QLIST_EMPTY_RCU(head) (atomic_rcu_read(&(head)->lh_first) == NULL)
+#define QLIST_FIRST_RCU(head) (atomic_rcu_read(&(head)->lh_first))
+#define QLIST_NEXT_RCU(elm, field) (atomic_rcu_read(&(elm)->field.le_next))
+
+/*
+ * List functions.
+ */
+
+
+/*
+ *  The difference between atomic_read/set and atomic_rcu_read/set
+ *  is in the including of a read/write memory barrier to the volatile
+ *  access. atomic_rcu_* macros include the memory barrier, the
+ *  plain atomic macros do not. Therefore, it should be correct to
+ *  issue a series of reads or writes to the same element using only
+ *  the atomic_* macro, until the last read or write, which should be
+ *  atomic_rcu_* to introduce a read or write memory barrier as
+ *  appropriate.
+ */
+
+/* Upon publication of the listelm->next value, list readers
+ * will see the new node when following next pointers from
+ * antecedent nodes, but may not see the new node when following
+ * prev pointers from subsequent nodes until after the RCU grace
+ * period expires.
+ * see linux/include/rculist.h __list_add_rcu(new, prev, next)
+ */
+#define QLIST_INSERT_AFTER_RCU(listelm, elm, field) do {    \
+    (elm)->field.le_next = (listelm)->field.le_next;        \
+    (elm)->field.le_prev = &(listelm)->field.le_next;       \
+    atomic_rcu_set(&(listelm)->field.le_next, (elm));       \
+    if ((elm)->field.le_next != NULL) {                     \
+       (elm)->field.le_next->field.le_prev =                \
+        &(elm)->field.le_next;                              \
+    }                                                       \
+} while (/*CONSTCOND*/0)
+
+/* Upon publication of the listelm->prev->next value, list
+ * readers will see the new element when following prev pointers
+ * from subsequent elements, but may not see the new element
+ * when following next pointers from antecedent elements
+ * until after the RCU grace period expires.
+ */
+#define QLIST_INSERT_BEFORE_RCU(listelm, elm, field) do {   \
+    (elm)->field.le_prev = (listelm)->field.le_prev;        \
+    (elm)->field.le_next = (listelm);                       \
+    atomic_rcu_set((listelm)->field.le_prev, (elm));        \
+    (listelm)->field.le_prev = &(elm)->field.le_next;       \
+} while (/*CONSTCOND*/0)
+
+/* Upon publication of the head->first value, list readers
+ * will see the new element when following the head, but may
+ * not see the new element when following prev pointers from
+ * subsequent elements until after the RCU grace period has
+ * expired.
+ */
+#define QLIST_INSERT_HEAD_RCU(head, elm, field) do {    \
+    (elm)->field.le_prev = &(head)->lh_first;           \
+    (elm)->field.le_next = (head)->lh_first;            \
+    atomic_rcu_set((&(head)->lh_first), (elm));         \
+    if ((elm)->field.le_next != NULL) {                 \
+       (elm)->field.le_next->field.le_prev =            \
+        &(elm)->field.le_next;                          \
+    }                                                   \
+} while (/*CONSTCOND*/0)
+
+
+/* prior to publication of the elm->prev->next value, some list
+ * readers may still see the removed element when following
+ * the antecedent's next pointer.
+ */
+#define QLIST_REMOVE_RCU(elm, field) do {           \
+    if ((elm)->field.le_next != NULL) {             \
+       (elm)->field.le_next->field.le_prev =        \
+        (elm)->field.le_prev;                       \
+    }                                               \
+    *(elm)->field.le_prev =  (elm)->field.le_next;  \
+} while (/*CONSTCOND*/0)
+
+/* List traversal must occur within an RCU critical section.  */
+#define QLIST_FOREACH_RCU(var, head, field)                 \
+        for ((var) = atomic_rcu_read(&(head)->lh_first);    \
+                (var);                                      \
+                (var) = atomic_rcu_read(&(var)->field.le_next))
+
+/* List traversal must occur within an RCU critical section.  */
+#define QLIST_FOREACH_SAFE_RCU(var, head, field, next_var)           \
+    for ((var) = (atomic_rcu_read(&(head)->lh_first));               \
+      (var) &&                                                       \
+          ((next_var) = atomic_rcu_read(&(var)->field.le_next), 1);  \
+           (var) = (next_var))
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* QEMU_RCU_QUEUE.H */
diff --git a/tests/Makefile b/tests/Makefile
index 06acfa1..a68ff89 100644
--- a/tests/Makefile
+++ b/tests/Makefile
@@ -62,6 +62,8 @@ check-unit-y += tests/test-int128$(EXESUF)
 gcov-files-test-int128-y =
 check-unit-y += tests/rcutorture$(EXESUF)
 gcov-files-rcutorture-y = util/rcu.c
+check-unit-y += tests/test-rcu-list$(EXESUF)
+gcov-files-test-rcu-list-y = util/rcu.c
 check-unit-y += tests/test-bitops$(EXESUF)
 check-unit-$(CONFIG_HAS_GLIB_SUBPROCESS_TESTS) += tests/test-qdev-global-props$(EXESUF)
 check-unit-y += tests/check-qom-interface$(EXESUF)
@@ -226,7 +228,7 @@ test-obj-y = tests/check-qint.o tests/check-qstring.o tests/check-qdict.o \
 	tests/test-qmp-commands.o tests/test-visitor-serialization.o \
 	tests/test-x86-cpuid.o tests/test-mul64.o tests/test-int128.o \
 	tests/test-opts-visitor.o tests/test-qmp-event.o \
-	tests/rcutorture.o
+	tests/rcutorture.o tests/test-rcu-list.o
 
 test-qapi-obj-y = tests/test-qapi-visit.o tests/test-qapi-types.o \
 		  tests/test-qapi-event.o
@@ -256,6 +258,7 @@ tests/test-xbzrle$(EXESUF): tests/test-xbzrle.o migration/xbzrle.o page_cache.o
 tests/test-cutils$(EXESUF): tests/test-cutils.o util/cutils.o
 tests/test-int128$(EXESUF): tests/test-int128.o
 tests/rcutorture$(EXESUF): tests/rcutorture.o libqemuutil.a libqemustub.a
+tests/test-rcu-list$(EXESUF): tests/test-rcu-list.o libqemuutil.a libqemustub.a
 
 tests/test-qdev-global-props$(EXESUF): tests/test-qdev-global-props.o \
 	hw/core/qdev.o hw/core/qdev-properties.o hw/core/hotplug.o\
diff --git a/tests/test-rcu-list.c b/tests/test-rcu-list.c
new file mode 100644
index 0000000..46b5e26
--- /dev/null
+++ b/tests/test-rcu-list.c
@@ -0,0 +1,306 @@
+/*
+ * rcuq_test.c
+ *
+ * usage: rcuq_test <readers> <duration>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (c) 2013 Mike D. Day, IBM Corporation.
+ */
+
+#include <glib.h>
+#include <stdlib.h>
+#include <stdio.h>
+#include <string.h>
+#include "qemu/atomic.h"
+#include "qemu/rcu.h"
+#include "qemu/compiler.h"
+#include "qemu/osdep.h"
+#include "qemu/thread.h"
+#include "qemu/rcu_queue.h"
+
+/*
+ * Test variables.
+ */
+
+long long n_reads = 0LL;
+long long n_updates = 0LL;
+long long n_reclaims = 0LL;
+long long n_nodes_removed = 0LL;
+long long n_nodes = 0LL;
+int g_test_in_charge = 0;
+
+int nthreadsrunning;
+
+char argsbuf[64];
+
+#define GOFLAG_INIT 0
+#define GOFLAG_RUN  1
+#define GOFLAG_STOP 2
+
+static volatile int goflag = GOFLAG_INIT;
+
+#define RCU_READ_RUN 1000
+#define RCU_UPDATE_RUN 10
+#define NR_THREADS 100
+#define RCU_Q_LEN 100
+
+static QemuThread threads[NR_THREADS];
+static struct rcu_reader_data *data[NR_THREADS];
+static int n_threads;
+
+static int select_random_el(int max)
+{
+    return (rand() % max);
+}
+
+
+static void create_thread(void *(*func)(void *))
+{
+    if (n_threads >= NR_THREADS) {
+        fprintf(stderr, "Thread limit of %d exceeded!\n", NR_THREADS);
+        exit(-1);
+    }
+    qemu_thread_create(&threads[n_threads], "test", func, &data[n_threads],
+                       QEMU_THREAD_JOINABLE);
+    n_threads++;
+}
+
+static void wait_all_threads(void)
+{
+    int i;
+
+    for (i = 0; i < n_threads; i++) {
+        qemu_thread_join(&threads[i]);
+    }
+    n_threads = 0;
+}
+
+
+struct list_element {
+    QLIST_ENTRY(list_element) entry;
+    struct rcu_head rcu;
+    long long val;
+};
+
+static void reclaim_list_el(struct rcu_head *prcu)
+{
+    struct list_element *el = container_of(prcu, struct list_element, rcu);
+    g_free(el);
+    atomic_add(&n_reclaims, 1);
+}
+
+static QLIST_HEAD(q_list_head, list_element) Q_list_head;
+
+static void *rcu_q_reader(void *arg)
+{
+    long long j, n_reads_local = 0;
+    struct list_element *el;
+
+    *(struct rcu_reader_data **)arg = &rcu_reader;
+    atomic_inc(&nthreadsrunning);
+    while (goflag == GOFLAG_INIT) {
+        g_usleep(1000);
+    }
+
+    while (goflag == GOFLAG_RUN) {
+        rcu_read_lock();
+        QLIST_FOREACH_RCU(el, &Q_list_head, entry) {
+            j = atomic_read(&el->val);
+            (void)j;
+            n_reads_local++;
+            if (goflag == GOFLAG_STOP) {
+                break;
+            }
+        }
+        rcu_read_unlock();
+
+        g_usleep(100);
+    }
+    atomic_add(&n_reads, n_reads_local);
+    return NULL;
+}
+
+
+static void *rcu_q_updater(void *arg)
+{
+    int j, target_el;
+    long long n_updates_local = 0;
+    long long n_removed_local = 0;
+    struct list_element *el, *prev_el;
+
+    *(struct rcu_reader_data **)arg = &rcu_reader;
+    atomic_inc(&nthreadsrunning);
+    while (goflag == GOFLAG_INIT) {
+        g_usleep(1000);
+    }
+
+    while (goflag == GOFLAG_RUN) {
+        target_el = select_random_el(RCU_Q_LEN);
+        j = 0;
+        /* FOREACH_RCU could work here but let's use both macros */
+        QLIST_FOREACH_SAFE_RCU(prev_el, &Q_list_head, entry, el) {
+            j++;
+            if (target_el == j) {
+                QLIST_REMOVE_RCU(prev_el, entry);
+                /* may be more than one updater in the future */
+                call_rcu1(&prev_el->rcu, reclaim_list_el);
+                n_removed_local++;
+                break;
+            }
+        }
+        if (goflag == GOFLAG_STOP) {
+            break;
+        }
+        target_el = select_random_el(RCU_Q_LEN);
+        j = 0;
+        QLIST_FOREACH_RCU(el, &Q_list_head, entry) {
+            j++;
+            if (target_el == j) {
+                prev_el = g_new(struct list_element, 1);
+                atomic_add(&n_nodes, 1);
+                prev_el->val = atomic_read(&n_nodes);
+                QLIST_INSERT_BEFORE_RCU(el, prev_el, entry);
+                break;
+            }
+        }
+
+        n_updates_local += 2;
+        synchronize_rcu();
+    }
+    synchronize_rcu();
+    atomic_add(&n_updates, n_updates_local);
+    atomic_add(&n_nodes_removed, n_removed_local);
+    return NULL;
+}
+
+static void rcu_qtest_init(void)
+{
+    struct list_element *new_el;
+    int i;
+    nthreadsrunning = 0;
+    srand(time(0));
+    for (i = 0; i < RCU_Q_LEN; i++) {
+        new_el = g_new(struct list_element, 1);
+        new_el->val = i;
+        QLIST_INSERT_HEAD_RCU(&Q_list_head, new_el, entry);
+    }
+    atomic_add(&n_nodes, RCU_Q_LEN);
+}
+
+static void rcu_qtest_run(int duration, int nreaders)
+{
+    int nthreads = nreaders + 1;
+    while (atomic_read(&nthreadsrunning) < nthreads) {
+        g_usleep(1000);
+    }
+
+    goflag = GOFLAG_RUN;
+    sleep(duration);
+    goflag = GOFLAG_STOP;
+    wait_all_threads();
+}
+
+
+static void rcu_qtest(const char *test, int duration, int nreaders)
+{
+    int i;
+    long long n_removed_local = 0;
+
+    struct list_element *el, *prev_el;
+
+    rcu_qtest_init();
+    for (i = 0; i < nreaders; i++) {
+        create_thread(rcu_q_reader);
+    }
+    create_thread(rcu_q_updater);
+    rcu_qtest_run(duration, nreaders);
+
+    QLIST_FOREACH_SAFE_RCU(prev_el, &Q_list_head, entry, el) {
+        QLIST_REMOVE_RCU(prev_el, entry);
+        call_rcu1(&prev_el->rcu, reclaim_list_el);
+        n_removed_local++;
+    }
+    atomic_add(&n_nodes_removed, n_removed_local);
+    synchronize_rcu();
+    while (n_nodes_removed > n_reclaims) {
+        g_usleep(100);
+        synchronize_rcu();
+    }
+    if (g_test_in_charge) {
+        g_assert_cmpint(n_nodes_removed, ==, n_reclaims);
+    } else {
+        printf("%s: %d readers; 1 updater; nodes read: "  \
+               "%lld, nodes removed: %lld; nodes reclaimed: %lld\n",
+               test, nthreadsrunning - 1, n_reads, n_nodes_removed, n_reclaims);
+        exit(0);
+    }
+}
+
+static void usage(int argc, char *argv[])
+{
+    fprintf(stderr, "Usage: %s duration nreaders\n", argv[0]);
+    exit(-1);
+}
+
+static int gtest_seconds;
+
+static void gtest_rcuq_one(void)
+{
+    rcu_qtest("rcuqtest", gtest_seconds / 4, 1);
+}
+
+static void gtest_rcuq_few(void)
+{
+    rcu_qtest("rcuqtest", gtest_seconds / 4, 5);
+}
+
+static void gtest_rcuq_many(void)
+{
+    rcu_qtest("rcuqtest", gtest_seconds / 2, 20);
+}
+
+
+int main(int argc, char *argv[])
+{
+    int duration = 0, readers = 0;
+
+    if (argc >= 2) {
+        if (argv[1][0] == '-') {
+            g_test_init(&argc, &argv, NULL);
+            if (g_test_quick()) {
+                gtest_seconds = 4;
+            } else {
+                gtest_seconds = 20;
+            }
+            g_test_add_func("/rcu/qlist/single-threaded", gtest_rcuq_one);
+            g_test_add_func("/rcu/qlist/short-few", gtest_rcuq_few);
+            g_test_add_func("/rcu/qlist/long-many", gtest_rcuq_many);
+            g_test_in_charge = 1;
+            return g_test_run();
+        }
+        duration = strtoul(argv[1], NULL, 0);
+    }
+    if (argc >= 3) {
+        readers = strtoul(argv[2], NULL, 0);
+    }
+    if (duration && readers) {
+        rcu_qtest(argv[0], duration, readers);
+        return 0;
+    }
+
+    usage(argc, argv);
+    return -1;
+}
-- 
2.3.0

[Qemu-devel] [PULL 17/21] rcu: add g_free_rcu

[Paolo Bonzini]

This simplifies calling g_free from an RCU callback.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 docs/rcu.txt       | 11 +++++++----
 include/qemu/rcu.h |  8 ++++++++
 2 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/docs/rcu.txt b/docs/rcu.txt
index 61752b9..21ecb81 100644
--- a/docs/rcu.txt
+++ b/docs/rcu.txt
@@ -120,12 +120,15 @@ The core RCU API is small:
      void call_rcu(T *p,
                    void (*func)(T *p),
                    field-name);
+     void g_free_rcu(T *p,
+                     field-name);
 
-        call_rcu1 is typically used through this macro, in the common case
-        where the "struct rcu_head" is the first field in the struct.  In
-        the above case, one could have written simply:
+        call_rcu1 is typically used through these macro, in the common case
+        where the "struct rcu_head" is the first field in the struct.  If
+        the callback function is g_free, in particular, g_free_rcu can be
+        used.  In the above case, one could have written simply:
 
-            call_rcu(foo_reclaim, g_free, rcu);
+            g_free_rcu(foo_reclaim, rcu);
 
      typeof(*p) atomic_rcu_read(p);
 
diff --git a/include/qemu/rcu.h b/include/qemu/rcu.h
index 068a279..506ab58 100644
--- a/include/qemu/rcu.h
+++ b/include/qemu/rcu.h
@@ -140,6 +140,14 @@ extern void call_rcu1(struct rcu_head *head, RCUCBFunc *func);
       }),                                                                \
       (RCUCBFunc *)(func))
 
+#define g_free_rcu(obj, field) \
+    call_rcu1(({                                                         \
+        char __attribute__((unused))                                     \
+            offset_must_be_zero[-offsetof(typeof(*(obj)), field)];       \
+        &(obj)->field;                                                   \
+      }),                                                                \
+      (RCUCBFunc *)g_free);
+
 #ifdef __cplusplus
 }
 #endif
-- 
2.3.0

[Qemu-devel] [PULL 18/21] exec: protect mru_block with RCU

[Paolo Bonzini]

Hence, freeing a RAMBlock has to be switched to call_rcu.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 exec.c                 | 52 +++++++++++++++++++++++++++++++++++---------------
 include/exec/cpu-all.h |  2 ++
 2 files changed, 39 insertions(+), 15 deletions(-)

diff --git a/exec.c b/exec.c
index 76b3b6c..5a27c42 100644
--- a/exec.c
+++ b/exec.c
@@ -811,7 +811,7 @@ static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
     RAMBlock *block;
 
     /* The list is protected by the iothread lock here.  */
-    block = ram_list.mru_block;
+    block = atomic_rcu_read(&ram_list.mru_block);
     if (block && addr - block->offset < block->max_length) {
         goto found;
     }
@@ -825,6 +825,22 @@ static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
     abort();
 
 found:
+    /* It is safe to write mru_block outside the iothread lock.  This
+     * is what happens:
+     *
+     *     mru_block = xxx
+     *     rcu_read_unlock()
+     *                                        xxx removed from list
+     *                  rcu_read_lock()
+     *                  read mru_block
+     *                                        mru_block = NULL;
+     *                                        call_rcu(reclaim_ramblock, xxx);
+     *                  rcu_read_unlock()
+     *
+     * atomic_rcu_set is not needed here.  The block was already published
+     * when it was placed into the list.  Here we're just making an extra
+     * copy of the pointer.
+     */
     ram_list.mru_block = block;
     return block;
 }
@@ -1526,13 +1542,31 @@ void qemu_ram_free_from_ptr(ram_addr_t addr)
             QTAILQ_REMOVE(&ram_list.blocks, block, next);
             ram_list.mru_block = NULL;
             ram_list.version++;
-            g_free(block);
+            g_free_rcu(block, rcu);
             break;
         }
     }
     qemu_mutex_unlock_ramlist();
 }
 
+static void reclaim_ramblock(RAMBlock *block)
+{
+    if (block->flags & RAM_PREALLOC) {
+        ;
+    } else if (xen_enabled()) {
+        xen_invalidate_map_cache_entry(block->host);
+#ifndef _WIN32
+    } else if (block->fd >= 0) {
+        munmap(block->host, block->max_length);
+        close(block->fd);
+#endif
+    } else {
+        qemu_anon_ram_free(block->host, block->max_length);
+    }
+    g_free(block);
+}
+
+/* Called with the iothread lock held */
 void qemu_ram_free(ram_addr_t addr)
 {
     RAMBlock *block;
@@ -1544,19 +1578,7 @@ void qemu_ram_free(ram_addr_t addr)
             QTAILQ_REMOVE(&ram_list.blocks, block, next);
             ram_list.mru_block = NULL;
             ram_list.version++;
-            if (block->flags & RAM_PREALLOC) {
-                ;
-            } else if (xen_enabled()) {
-                xen_invalidate_map_cache_entry(block->host);
-#ifndef _WIN32
-            } else if (block->fd >= 0) {
-                munmap(block->host, block->max_length);
-                close(block->fd);
-#endif
-            } else {
-                qemu_anon_ram_free(block->host, block->max_length);
-            }
-            g_free(block);
+            call_rcu(block, reclaim_ramblock, rcu);
             break;
         }
     }
diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
index 2c48286..b8781d1 100644
--- a/include/exec/cpu-all.h
+++ b/include/exec/cpu-all.h
@@ -24,6 +24,7 @@
 #include "exec/memory.h"
 #include "qemu/thread.h"
 #include "qom/cpu.h"
+#include "qemu/rcu.h"
 
 /* some important defines:
  *
@@ -268,6 +269,7 @@ CPUArchState *cpu_copy(CPUArchState *env);
 typedef struct RAMBlock RAMBlock;
 
 struct RAMBlock {
+    struct rcu_head rcu;
     struct MemoryRegion *mr;
     uint8_t *host;
     ram_addr_t offset;
-- 
2.3.0

[Qemu-devel] [PULL 19/21] cosmetic changes preparing for the following patches

[Paolo Bonzini]

From: Mike Day <ncmike@ncultra.org>

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Mike Day <ncmike@ncultra.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch_init.c            |  9 +++---
 exec.c                 | 86 +++++++++++++++++++++++++++++++++-----------------
 include/exec/cpu-all.h |  1 +
 3 files changed, 62 insertions(+), 34 deletions(-)

diff --git a/arch_init.c b/arch_init.c
index 89c8fa4..71a539e 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -487,7 +487,6 @@ static void migration_bitmap_sync_range(ram_addr_t start, ram_addr_t length)
 }
 
 
-/* Needs iothread lock! */
 /* Fix me: there are too many global variables used in migration process. */
 static int64_t start_time;
 static int64_t bytes_xfer_prev;
@@ -500,6 +499,7 @@ static void migration_bitmap_sync_init(void)
     num_dirty_pages_period = 0;
 }
 
+/* Called with iothread lock held, to protect ram_list.dirty_memory[] */
 static void migration_bitmap_sync(void)
 {
     RAMBlock *block;
@@ -688,9 +688,9 @@ static int ram_find_and_save_block(QEMUFile *f, bool last_stage)
             }
         }
     }
+
     last_seen_block = block;
     last_offset = offset;
-
     return bytes_sent;
 }
 
@@ -816,6 +816,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
         acct_clear();
     }
 
+    /* iothread lock needed for ram_list.dirty_memory[] */
     qemu_mutex_lock_iothread();
     qemu_mutex_lock_ramlist();
     bytes_transferred = 0;
@@ -928,6 +929,7 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
     return total_sent;
 }
 
+/* Called with iothread lock */
 static int ram_save_complete(QEMUFile *f, void *opaque)
 {
     qemu_mutex_lock_ramlist();
@@ -1117,7 +1119,6 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
                 ret = -EINVAL;
                 break;
             }
-
             ch = qemu_get_byte(f);
             ram_handle_compressed(host, ch, TARGET_PAGE_SIZE);
             break;
@@ -1128,7 +1129,6 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
                 ret = -EINVAL;
                 break;
             }
-
             qemu_get_buffer(f, host, TARGET_PAGE_SIZE);
             break;
         case RAM_SAVE_FLAG_XBZRLE:
@@ -1138,7 +1138,6 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
                 ret = -EINVAL;
                 break;
             }
-
             if (load_xbzrle(f, addr, host) < 0) {
                 error_report("Failed to decompress XBZRLE page at "
                              RAM_ADDR_FMT, addr);
diff --git a/exec.c b/exec.c
index 5a27c42..2b6ee55 100644
--- a/exec.c
+++ b/exec.c
@@ -1265,11 +1265,12 @@ static RAMBlock *find_ram_block(ram_addr_t addr)
     return NULL;
 }
 
+/* Called with iothread lock held.  */
 void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
 {
-    RAMBlock *new_block = find_ram_block(addr);
-    RAMBlock *block;
+    RAMBlock *new_block, *block;
 
+    new_block = find_ram_block(addr);
     assert(new_block);
     assert(!new_block->idstr[0]);
 
@@ -1282,7 +1283,6 @@ void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
     }
     pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
 
-    /* This assumes the iothread lock is taken here too.  */
     qemu_mutex_lock_ramlist();
     QTAILQ_FOREACH(block, &ram_list.blocks, next) {
         if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
@@ -1294,10 +1294,17 @@ void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
     qemu_mutex_unlock_ramlist();
 }
 
+/* Called with iothread lock held.  */
 void qemu_ram_unset_idstr(ram_addr_t addr)
 {
-    RAMBlock *block = find_ram_block(addr);
+    RAMBlock *block;
 
+    /* FIXME: arch_init.c assumes that this is not called throughout
+     * migration.  Ignore the problem since hot-unplug during migration
+     * does not work anyway.
+     */
+
+    block = find_ram_block(addr);
     if (block) {
         memset(block->idstr, 0, sizeof(block->idstr));
     }
@@ -1405,6 +1412,8 @@ static ram_addr_t ram_block_add(RAMBlock *new_block, Error **errp)
 
     if (new_ram_size > old_ram_size) {
         int i;
+
+        /* ram_list.dirty_memory[] is protected by the iothread lock.  */
         for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
             ram_list.dirty_memory[i] =
                 bitmap_zero_extend(ram_list.dirty_memory[i],
@@ -1583,7 +1592,6 @@ void qemu_ram_free(ram_addr_t addr)
         }
     }
     qemu_mutex_unlock_ramlist();
-
 }
 
 #ifndef _WIN32
@@ -1631,7 +1639,6 @@ void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
                 memory_try_enable_merging(vaddr, length);
                 qemu_ram_setup_dump(vaddr, length);
             }
-            return;
         }
     }
 }
@@ -1639,49 +1646,60 @@ void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
 
 int qemu_get_ram_fd(ram_addr_t addr)
 {
-    RAMBlock *block = qemu_get_ram_block(addr);
+    RAMBlock *block;
+    int fd;
 
-    return block->fd;
+    block = qemu_get_ram_block(addr);
+    fd = block->fd;
+    return fd;
 }
 
 void *qemu_get_ram_block_host_ptr(ram_addr_t addr)
 {
-    RAMBlock *block = qemu_get_ram_block(addr);
+    RAMBlock *block;
+    void *ptr;
 
-    return ramblock_ptr(block, 0);
+    block = qemu_get_ram_block(addr);
+    ptr = ramblock_ptr(block, 0);
+    return ptr;
 }
 
 /* Return a host pointer to ram allocated with qemu_ram_alloc.
-   With the exception of the softmmu code in this file, this should
-   only be used for local memory (e.g. video ram) that the device owns,
-   and knows it isn't going to access beyond the end of the block.
-
-   It should not be used for general purpose DMA.
-   Use cpu_physical_memory_map/cpu_physical_memory_rw instead.
+ * This should not be used for general purpose DMA.  Use address_space_map
+ * or address_space_rw instead. For local memory (e.g. video ram) that the
+ * device owns, use memory_region_get_ram_ptr.
  */
 void *qemu_get_ram_ptr(ram_addr_t addr)
 {
-    RAMBlock *block = qemu_get_ram_block(addr);
+    RAMBlock *block;
+    void *ptr;
 
-    if (xen_enabled()) {
+    block = qemu_get_ram_block(addr);
+
+    if (xen_enabled() && block->host == NULL) {
         /* We need to check if the requested address is in the RAM
          * because we don't want to map the entire memory in QEMU.
          * In that case just map until the end of the page.
          */
         if (block->offset == 0) {
-            return xen_map_cache(addr, 0, 0);
-        } else if (block->host == NULL) {
-            block->host =
-                xen_map_cache(block->offset, block->max_length, 1);
+            ptr = xen_map_cache(addr, 0, 0);
+            goto done;
         }
+
+        block->host = xen_map_cache(block->offset, block->max_length, 1);
     }
-    return ramblock_ptr(block, addr - block->offset);
+    ptr = ramblock_ptr(block, addr - block->offset);
+
+done:
+    return ptr;
 }
 
 /* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr
- * but takes a size argument */
+ * but takes a size argument.
+ */
 static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
 {
+    void *ptr;
     if (*size == 0) {
         return NULL;
     }
@@ -1689,12 +1707,12 @@ static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
         return xen_map_cache(addr, *size, 1);
     } else {
         RAMBlock *block;
-
         QTAILQ_FOREACH(block, &ram_list.blocks, next) {
             if (addr - block->offset < block->max_length) {
                 if (addr - block->offset + *size > block->max_length)
                     *size = block->max_length - addr + block->offset;
-                return ramblock_ptr(block, addr - block->offset);
+                ptr = ramblock_ptr(block, addr - block->offset);
+                return ptr;
             }
         }
 
@@ -1704,15 +1722,24 @@ static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
 }
 
 /* Some of the softmmu routines need to translate from a host pointer
-   (typically a TLB entry) back to a ram offset.  */
+ * (typically a TLB entry) back to a ram offset.
+ *
+ * By the time this function returns, the returned pointer is not protected
+ * by RCU anymore.  If the caller is not within an RCU critical section and
+ * does not hold the iothread lock, it must have other means of protecting the
+ * pointer, such as a reference to the region that includes the incoming
+ * ram_addr_t.
+ */
 MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
 {
     RAMBlock *block;
     uint8_t *host = ptr;
+    MemoryRegion *mr;
 
     if (xen_enabled()) {
         *ram_addr = xen_ram_addr_from_mapcache(ptr);
-        return qemu_get_ram_block(*ram_addr)->mr;
+        mr = qemu_get_ram_block(*ram_addr)->mr;
+        return mr;
     }
 
     block = ram_list.mru_block;
@@ -1734,7 +1761,8 @@ MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
 
 found:
     *ram_addr = block->offset + (host - block->host);
-    return block->mr;
+    mr = block->mr;
+    return mr;
 }
 
 static void notdirty_mem_write(void *opaque, hwaddr ram_addr,
diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
index b8781d1..f7a3625 100644
--- a/include/exec/cpu-all.h
+++ b/include/exec/cpu-all.h
@@ -277,6 +277,7 @@ struct RAMBlock {
     ram_addr_t max_length;
     void (*resized)(const char*, uint64_t length, void *host);
     uint32_t flags;
+    /* Protected by iothread lock.  */
     char idstr[256];
     /* Reads can take either the iothread or the ramlist lock.
      * Writes must take both locks.
-- 
2.3.0

[Qemu-devel] [PULL 20/21] exec: convert ram_list to QLIST

[Paolo Bonzini]

From: Mike Day <ncmike@ncultra.org>

QLIST has RCU-friendly primitives, so switch to it.

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Mike Day <ncmike@ncultra.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch_init.c                  | 19 ++++++++--------
 exec.c                       | 52 +++++++++++++++++++++++++-------------------
 include/exec/cpu-all.h       |  4 ++--
 scripts/dump-guest-memory.py |  8 +++----
 4 files changed, 46 insertions(+), 37 deletions(-)

diff --git a/arch_init.c b/arch_init.c
index 71a539e..1ee2e35 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -523,7 +523,7 @@ static void migration_bitmap_sync(void)
     trace_migration_bitmap_sync_start();
     address_space_sync_dirty_bitmap(&address_space_memory);
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         migration_bitmap_sync_range(block->mr->ram_addr, block->used_length);
     }
     trace_migration_bitmap_sync_end(migration_dirty_pages
@@ -661,7 +661,7 @@ static int ram_find_and_save_block(QEMUFile *f, bool last_stage)
     MemoryRegion *mr;
 
     if (!block)
-        block = QTAILQ_FIRST(&ram_list.blocks);
+        block = QLIST_FIRST(&ram_list.blocks);
 
     while (true) {
         mr = block->mr;
@@ -672,9 +672,9 @@ static int ram_find_and_save_block(QEMUFile *f, bool last_stage)
         }
         if (offset >= block->used_length) {
             offset = 0;
-            block = QTAILQ_NEXT(block, next);
+            block = QLIST_NEXT(block, next);
             if (!block) {
-                block = QTAILQ_FIRST(&ram_list.blocks);
+                block = QLIST_FIRST(&ram_list.blocks);
                 complete_round = true;
                 ram_bulk_stage = false;
             }
@@ -728,8 +728,9 @@ uint64_t ram_bytes_total(void)
     RAMBlock *block;
     uint64_t total = 0;
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next)
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         total += block->used_length;
+    }
 
     return total;
 }
@@ -831,7 +832,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
      * gaps due to alignment or unplugs.
      */
     migration_dirty_pages = 0;
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         uint64_t block_pages;
 
         block_pages = block->used_length >> TARGET_PAGE_BITS;
@@ -844,7 +845,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
 
     qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         qemu_put_byte(f, strlen(block->idstr));
         qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
         qemu_put_be64(f, block->used_length);
@@ -1031,7 +1032,7 @@ static inline void *host_from_stream_offset(QEMUFile *f,
     qemu_get_buffer(f, (uint8_t *)id, len);
     id[len] = 0;
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         if (!strncmp(id, block->idstr, sizeof(id)) &&
             block->max_length > offset) {
             return memory_region_get_ram_ptr(block->mr) + offset;
@@ -1088,7 +1089,7 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
                 id[len] = 0;
                 length = qemu_get_be64(f);
 
-                QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+                QLIST_FOREACH(block, &ram_list.blocks, next) {
                     if (!strncmp(id, block->idstr, sizeof(id))) {
                         if (length != block->used_length) {
                             Error *local_err = NULL;
diff --git a/exec.c b/exec.c
index 2b6ee55..a7c2b92 100644
--- a/exec.c
+++ b/exec.c
@@ -58,7 +58,7 @@
 #if !defined(CONFIG_USER_ONLY)
 static bool in_migration;
 
-RAMList ram_list = { .blocks = QTAILQ_HEAD_INITIALIZER(ram_list.blocks) };
+RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) };
 
 static MemoryRegion *system_memory;
 static MemoryRegion *system_io;
@@ -815,7 +815,7 @@ static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
     if (block && addr - block->offset < block->max_length) {
         goto found;
     }
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         if (addr - block->offset < block->max_length) {
             goto found;
         }
@@ -1197,15 +1197,16 @@ static ram_addr_t find_ram_offset(ram_addr_t size)
 
     assert(size != 0); /* it would hand out same offset multiple times */
 
-    if (QTAILQ_EMPTY(&ram_list.blocks))
+    if (QLIST_EMPTY(&ram_list.blocks)) {
         return 0;
+    }
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         ram_addr_t end, next = RAM_ADDR_MAX;
 
         end = block->offset + block->max_length;
 
-        QTAILQ_FOREACH(next_block, &ram_list.blocks, next) {
+        QLIST_FOREACH(next_block, &ram_list.blocks, next) {
             if (next_block->offset >= end) {
                 next = MIN(next, next_block->offset);
             }
@@ -1230,9 +1231,9 @@ ram_addr_t last_ram_offset(void)
     RAMBlock *block;
     ram_addr_t last = 0;
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next)
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         last = MAX(last, block->offset + block->max_length);
-
+    }
     return last;
 }
 
@@ -1256,7 +1257,7 @@ static RAMBlock *find_ram_block(ram_addr_t addr)
 {
     RAMBlock *block;
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         if (block->offset == addr) {
             return block;
         }
@@ -1284,7 +1285,7 @@ void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
     pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
 
     qemu_mutex_lock_ramlist();
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
             fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
                     new_block->idstr);
@@ -1366,6 +1367,7 @@ int qemu_ram_resize(ram_addr_t base, ram_addr_t newsize, Error **errp)
 static ram_addr_t ram_block_add(RAMBlock *new_block, Error **errp)
 {
     RAMBlock *block;
+    RAMBlock *last_block = NULL;
     ram_addr_t old_ram_size, new_ram_size;
 
     old_ram_size = last_ram_offset() >> TARGET_PAGE_BITS;
@@ -1392,16 +1394,22 @@ static ram_addr_t ram_block_add(RAMBlock *new_block, Error **errp)
         }
     }
 
-    /* Keep the list sorted from biggest to smallest block.  */
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    /* Keep the list sorted from biggest to smallest block.  Unlike QTAILQ,
+     * QLIST (which has an RCU-friendly variant) does not have insertion at
+     * tail, so save the last element in last_block.
+     */
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
+        last_block = block;
         if (block->max_length < new_block->max_length) {
             break;
         }
     }
     if (block) {
-        QTAILQ_INSERT_BEFORE(block, new_block, next);
-    } else {
-        QTAILQ_INSERT_TAIL(&ram_list.blocks, new_block, next);
+        QLIST_INSERT_BEFORE(block, new_block, next);
+    } else if (last_block) {
+        QLIST_INSERT_AFTER(last_block, new_block, next);
+    } else { /* list is empty */
+        QLIST_INSERT_HEAD(&ram_list.blocks, new_block, next);
     }
     ram_list.mru_block = NULL;
 
@@ -1546,9 +1554,9 @@ void qemu_ram_free_from_ptr(ram_addr_t addr)
 
     /* This assumes the iothread lock is taken here too.  */
     qemu_mutex_lock_ramlist();
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         if (addr == block->offset) {
-            QTAILQ_REMOVE(&ram_list.blocks, block, next);
+            QLIST_REMOVE(block, next);
             ram_list.mru_block = NULL;
             ram_list.version++;
             g_free_rcu(block, rcu);
@@ -1582,9 +1590,9 @@ void qemu_ram_free(ram_addr_t addr)
 
     /* This assumes the iothread lock is taken here too.  */
     qemu_mutex_lock_ramlist();
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         if (addr == block->offset) {
-            QTAILQ_REMOVE(&ram_list.blocks, block, next);
+            QLIST_REMOVE(block, next);
             ram_list.mru_block = NULL;
             ram_list.version++;
             call_rcu(block, reclaim_ramblock, rcu);
@@ -1602,7 +1610,7 @@ void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
     int flags;
     void *area, *vaddr;
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         offset = addr - block->offset;
         if (offset < block->max_length) {
             vaddr = ramblock_ptr(block, offset);
@@ -1707,7 +1715,7 @@ static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
         return xen_map_cache(addr, *size, 1);
     } else {
         RAMBlock *block;
-        QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+        QLIST_FOREACH(block, &ram_list.blocks, next) {
             if (addr - block->offset < block->max_length) {
                 if (addr - block->offset + *size > block->max_length)
                     *size = block->max_length - addr + block->offset;
@@ -1747,7 +1755,7 @@ MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
         goto found;
     }
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         /* This case append when the block is not mapped. */
         if (block->host == NULL) {
             continue;
@@ -3019,7 +3027,7 @@ void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
 {
     RAMBlock *block;
 
-    QTAILQ_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH(block, &ram_list.blocks, next) {
         func(block->host, block->offset, block->used_length, opaque);
     }
 }
diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
index f7a3625..87b8658 100644
--- a/include/exec/cpu-all.h
+++ b/include/exec/cpu-all.h
@@ -282,7 +282,7 @@ struct RAMBlock {
     /* Reads can take either the iothread or the ramlist lock.
      * Writes must take both locks.
      */
-    QTAILQ_ENTRY(RAMBlock) next;
+    QLIST_ENTRY(RAMBlock) next;
     int fd;
 };
 
@@ -299,7 +299,7 @@ typedef struct RAMList {
     unsigned long *dirty_memory[DIRTY_MEMORY_NUM];
     RAMBlock *mru_block;
     /* Protected by the ramlist lock.  */
-    QTAILQ_HEAD(, RAMBlock) blocks;
+    QLIST_HEAD(, RAMBlock) blocks;
     uint32_t version;
 } RAMList;
 extern RAMList ram_list;
diff --git a/scripts/dump-guest-memory.py b/scripts/dump-guest-memory.py
index 1ed8b67..dc8e44a 100644
--- a/scripts/dump-guest-memory.py
+++ b/scripts/dump-guest-memory.py
@@ -108,16 +108,16 @@ shape and this command should mostly work."""
         assert (val["hi"] == 0)
         return val["lo"]
 
-    def qtailq_foreach(self, head, field_str):
-        var_p = head["tqh_first"]
+    def qlist_foreach(self, head, field_str):
+        var_p = head["lh_first"]
         while (var_p != 0):
             var = var_p.dereference()
             yield var
-            var_p = var[field_str]["tqe_next"]
+            var_p = var[field_str]["le_next"]
 
     def qemu_get_ram_block(self, ram_addr):
         ram_blocks = gdb.parse_and_eval("ram_list.blocks")
-        for block in self.qtailq_foreach(ram_blocks, "next"):
+        for block in self.qlist_foreach(ram_blocks, "next"):
             if (ram_addr - block["offset"] < block["length"]):
                 return block
         raise gdb.GdbError("Bad ram offset %x" % ram_addr)
-- 
2.3.0

[Qemu-devel] [PULL 21/21] Convert ram_list to RCU

[Paolo Bonzini]

From: Mike Day <ncmike@ncultra.org>

Allow "unlocked" reads of the ram_list by using an RCU-enabled QLIST.

The ramlist mutex is kept.  call_rcu callbacks are run with the iothread
lock taken, but that may change in the future.  Writers still take the
ramlist mutex, but they no longer need to assume that the iothread lock
is taken.

Readers of the list, instead, no longer require either the iothread
or ramlist mutex, but they need to use rcu_read_lock() and
rcu_read_unlock().

One place in arch_init.c was downgrading from write side to read side
like this:

    qemu_mutex_lock_iothread()
    qemu_mutex_lock_ramlist()
    ...
    qemu_mutex_unlock_iothread()
    ...
    qemu_mutex_unlock_ramlist()

and the equivalent idiom is:

    qemu_mutex_lock_ramlist()
    rcu_read_lock()
    ...
    qemu_mutex_unlock_ramlist()
    ...
    rcu_read_unlock()

Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Mike Day <ncmike@ncultra.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
 arch_init.c            |  65 ++++++++++++++++++++++---------
 exec.c                 | 102 ++++++++++++++++++++++++++++++++++---------------
 include/exec/cpu-all.h |   6 +--
 3 files changed, 121 insertions(+), 52 deletions(-)

diff --git a/arch_init.c b/arch_init.c
index 1ee2e35..5fc6fc3 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -52,6 +52,7 @@
 #include "exec/ram_addr.h"
 #include "hw/acpi/acpi.h"
 #include "qemu/host-utils.h"
+#include "qemu/rcu_queue.h"
 
 #ifdef DEBUG_ARCH_INIT
 #define DPRINTF(fmt, ...) \
@@ -523,9 +524,12 @@ static void migration_bitmap_sync(void)
     trace_migration_bitmap_sync_start();
     address_space_sync_dirty_bitmap(&address_space_memory);
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    rcu_read_lock();
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         migration_bitmap_sync_range(block->mr->ram_addr, block->used_length);
     }
+    rcu_read_unlock();
+
     trace_migration_bitmap_sync_end(migration_dirty_pages
                                     - num_dirty_pages_init);
     num_dirty_pages_period += migration_dirty_pages - num_dirty_pages_init;
@@ -648,6 +652,8 @@ static int ram_save_page(QEMUFile *f, RAMBlock* block, ram_addr_t offset,
 /*
  * ram_find_and_save_block: Finds a page to send and sends it to f
  *
+ * Called within an RCU critical section.
+ *
  * Returns:  The number of bytes written.
  *           0 means no dirty pages
  */
@@ -661,7 +667,7 @@ static int ram_find_and_save_block(QEMUFile *f, bool last_stage)
     MemoryRegion *mr;
 
     if (!block)
-        block = QLIST_FIRST(&ram_list.blocks);
+        block = QLIST_FIRST_RCU(&ram_list.blocks);
 
     while (true) {
         mr = block->mr;
@@ -672,9 +678,9 @@ static int ram_find_and_save_block(QEMUFile *f, bool last_stage)
         }
         if (offset >= block->used_length) {
             offset = 0;
-            block = QLIST_NEXT(block, next);
+            block = QLIST_NEXT_RCU(block, next);
             if (!block) {
-                block = QLIST_FIRST(&ram_list.blocks);
+                block = QLIST_FIRST_RCU(&ram_list.blocks);
                 complete_round = true;
                 ram_bulk_stage = false;
             }
@@ -728,10 +734,10 @@ uint64_t ram_bytes_total(void)
     RAMBlock *block;
     uint64_t total = 0;
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    rcu_read_lock();
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next)
         total += block->used_length;
-    }
-
+    rcu_read_unlock();
     return total;
 }
 
@@ -777,6 +783,13 @@ static void reset_ram_globals(void)
 
 #define MAX_WAIT 50 /* ms, half buffered_file limit */
 
+
+/* Each of ram_save_setup, ram_save_iterate and ram_save_complete has
+ * long-running RCU critical section.  When rcu-reclaims in the code
+ * start to become numerous it will be necessary to reduce the
+ * granularity of these critical sections.
+ */
+
 static int ram_save_setup(QEMUFile *f, void *opaque)
 {
     RAMBlock *block;
@@ -820,6 +833,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
     /* iothread lock needed for ram_list.dirty_memory[] */
     qemu_mutex_lock_iothread();
     qemu_mutex_lock_ramlist();
+    rcu_read_lock();
     bytes_transferred = 0;
     reset_ram_globals();
 
@@ -832,7 +846,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
      * gaps due to alignment or unplugs.
      */
     migration_dirty_pages = 0;
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         uint64_t block_pages;
 
         block_pages = block->used_length >> TARGET_PAGE_BITS;
@@ -841,17 +855,18 @@ static int ram_save_setup(QEMUFile *f, void *opaque)
 
     memory_global_dirty_log_start();
     migration_bitmap_sync();
+    qemu_mutex_unlock_ramlist();
     qemu_mutex_unlock_iothread();
 
     qemu_put_be64(f, ram_bytes_total() | RAM_SAVE_FLAG_MEM_SIZE);
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         qemu_put_byte(f, strlen(block->idstr));
         qemu_put_buffer(f, (uint8_t *)block->idstr, strlen(block->idstr));
         qemu_put_be64(f, block->used_length);
     }
 
-    qemu_mutex_unlock_ramlist();
+    rcu_read_unlock();
 
     ram_control_before_iterate(f, RAM_CONTROL_SETUP);
     ram_control_after_iterate(f, RAM_CONTROL_SETUP);
@@ -868,12 +883,14 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
     int64_t t0;
     int total_sent = 0;
 
-    qemu_mutex_lock_ramlist();
-
+    rcu_read_lock();
     if (ram_list.version != last_version) {
         reset_ram_globals();
     }
 
+    /* Read version before ram_list.blocks */
+    smp_rmb();
+
     ram_control_before_iterate(f, RAM_CONTROL_ROUND);
 
     t0 = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
@@ -904,8 +921,7 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
         }
         i++;
     }
-
-    qemu_mutex_unlock_ramlist();
+    rcu_read_unlock();
 
     /*
      * Must occur before EOS (or any QEMUFile operation)
@@ -933,7 +949,8 @@ static int ram_save_iterate(QEMUFile *f, void *opaque)
 /* Called with iothread lock */
 static int ram_save_complete(QEMUFile *f, void *opaque)
 {
-    qemu_mutex_lock_ramlist();
+    rcu_read_lock();
+
     migration_bitmap_sync();
 
     ram_control_before_iterate(f, RAM_CONTROL_FINISH);
@@ -955,7 +972,7 @@ static int ram_save_complete(QEMUFile *f, void *opaque)
     ram_control_after_iterate(f, RAM_CONTROL_FINISH);
     migration_end();
 
-    qemu_mutex_unlock_ramlist();
+    rcu_read_unlock();
     qemu_put_be64(f, RAM_SAVE_FLAG_EOS);
 
     return 0;
@@ -969,7 +986,9 @@ static uint64_t ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size)
 
     if (remaining_size < max_size) {
         qemu_mutex_lock_iothread();
+        rcu_read_lock();
         migration_bitmap_sync();
+        rcu_read_unlock();
         qemu_mutex_unlock_iothread();
         remaining_size = ram_save_remaining() * TARGET_PAGE_SIZE;
     }
@@ -1011,6 +1030,9 @@ static int load_xbzrle(QEMUFile *f, ram_addr_t addr, void *host)
     return 0;
 }
 
+/* Must be called from within a rcu critical section.
+ * Returns a pointer from within the RCU-protected ram_list.
+ */
 static inline void *host_from_stream_offset(QEMUFile *f,
                                             ram_addr_t offset,
                                             int flags)
@@ -1032,7 +1054,7 @@ static inline void *host_from_stream_offset(QEMUFile *f,
     qemu_get_buffer(f, (uint8_t *)id, len);
     id[len] = 0;
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         if (!strncmp(id, block->idstr, sizeof(id)) &&
             block->max_length > offset) {
             return memory_region_get_ram_ptr(block->mr) + offset;
@@ -1065,6 +1087,12 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
         ret = -EINVAL;
     }
 
+    /* This RCU critical section can be very long running.
+     * When RCU reclaims in the code start to become numerous,
+     * it will be necessary to reduce the granularity of this
+     * critical section.
+     */
+    rcu_read_lock();
     while (!ret && !(flags & RAM_SAVE_FLAG_EOS)) {
         ram_addr_t addr, total_ram_bytes;
         void *host;
@@ -1089,7 +1117,7 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
                 id[len] = 0;
                 length = qemu_get_be64(f);
 
-                QLIST_FOREACH(block, &ram_list.blocks, next) {
+                QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
                     if (!strncmp(id, block->idstr, sizeof(id))) {
                         if (length != block->used_length) {
                             Error *local_err = NULL;
@@ -1163,6 +1191,7 @@ static int ram_load(QEMUFile *f, void *opaque, int version_id)
         }
     }
 
+    rcu_read_unlock();
     DPRINTF("Completed load of VM with exit code %d seq iteration "
             "%" PRIu64 "\n", ret, seq_iter);
     return ret;
diff --git a/exec.c b/exec.c
index a7c2b92..c85321a 100644
--- a/exec.c
+++ b/exec.c
@@ -44,7 +44,7 @@
 #include "trace.h"
 #endif
 #include "exec/cpu-all.h"
-
+#include "qemu/rcu_queue.h"
 #include "exec/cputlb.h"
 #include "translate-all.h"
 
@@ -58,6 +58,9 @@
 #if !defined(CONFIG_USER_ONLY)
 static bool in_migration;
 
+/* ram_list is read under rcu_read_lock()/rcu_read_unlock().  Writes
+ * are protected by the ramlist lock.
+ */
 RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) };
 
 static MemoryRegion *system_memory;
@@ -806,16 +809,16 @@ void cpu_abort(CPUState *cpu, const char *fmt, ...)
 }
 
 #if !defined(CONFIG_USER_ONLY)
+/* Called from RCU critical section */
 static RAMBlock *qemu_get_ram_block(ram_addr_t addr)
 {
     RAMBlock *block;
 
-    /* The list is protected by the iothread lock here.  */
     block = atomic_rcu_read(&ram_list.mru_block);
     if (block && addr - block->offset < block->max_length) {
         goto found;
     }
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         if (addr - block->offset < block->max_length) {
             goto found;
         }
@@ -854,10 +857,12 @@ static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length)
     end = TARGET_PAGE_ALIGN(start + length);
     start &= TARGET_PAGE_MASK;
 
+    rcu_read_lock();
     block = qemu_get_ram_block(start);
     assert(block == qemu_get_ram_block(end - 1));
     start1 = (uintptr_t)ramblock_ptr(block, start - block->offset);
     cpu_tlb_reset_dirty_all(start1, length);
+    rcu_read_unlock();
 }
 
 /* Note: start and end must be within the same ram block.  */
@@ -1190,6 +1195,7 @@ error:
 }
 #endif
 
+/* Called with the ramlist lock held.  */
 static ram_addr_t find_ram_offset(ram_addr_t size)
 {
     RAMBlock *block, *next_block;
@@ -1197,16 +1203,16 @@ static ram_addr_t find_ram_offset(ram_addr_t size)
 
     assert(size != 0); /* it would hand out same offset multiple times */
 
-    if (QLIST_EMPTY(&ram_list.blocks)) {
+    if (QLIST_EMPTY_RCU(&ram_list.blocks)) {
         return 0;
     }
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         ram_addr_t end, next = RAM_ADDR_MAX;
 
         end = block->offset + block->max_length;
 
-        QLIST_FOREACH(next_block, &ram_list.blocks, next) {
+        QLIST_FOREACH_RCU(next_block, &ram_list.blocks, next) {
             if (next_block->offset >= end) {
                 next = MIN(next, next_block->offset);
             }
@@ -1231,9 +1237,11 @@ ram_addr_t last_ram_offset(void)
     RAMBlock *block;
     ram_addr_t last = 0;
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    rcu_read_lock();
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         last = MAX(last, block->offset + block->max_length);
     }
+    rcu_read_unlock();
     return last;
 }
 
@@ -1253,11 +1261,14 @@ static void qemu_ram_setup_dump(void *addr, ram_addr_t size)
     }
 }
 
+/* Called within an RCU critical section, or while the ramlist lock
+ * is held.
+ */
 static RAMBlock *find_ram_block(ram_addr_t addr)
 {
     RAMBlock *block;
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         if (block->offset == addr) {
             return block;
         }
@@ -1271,6 +1282,7 @@ void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
 {
     RAMBlock *new_block, *block;
 
+    rcu_read_lock();
     new_block = find_ram_block(addr);
     assert(new_block);
     assert(!new_block->idstr[0]);
@@ -1284,15 +1296,14 @@ void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev)
     }
     pstrcat(new_block->idstr, sizeof(new_block->idstr), name);
 
-    qemu_mutex_lock_ramlist();
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         if (block != new_block && !strcmp(block->idstr, new_block->idstr)) {
             fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n",
                     new_block->idstr);
             abort();
         }
     }
-    qemu_mutex_unlock_ramlist();
+    rcu_read_unlock();
 }
 
 /* Called with iothread lock held.  */
@@ -1305,10 +1316,12 @@ void qemu_ram_unset_idstr(ram_addr_t addr)
      * does not work anyway.
      */
 
+    rcu_read_lock();
     block = find_ram_block(addr);
     if (block) {
         memset(block->idstr, 0, sizeof(block->idstr));
     }
+    rcu_read_unlock();
 }
 
 static int memory_try_enable_merging(void *addr, size_t len)
@@ -1372,7 +1385,6 @@ static ram_addr_t ram_block_add(RAMBlock *new_block, Error **errp)
 
     old_ram_size = last_ram_offset() >> TARGET_PAGE_BITS;
 
-    /* This assumes the iothread lock is taken here too.  */
     qemu_mutex_lock_ramlist();
     new_block->offset = find_ram_offset(new_block->max_length);
 
@@ -1398,21 +1410,23 @@ static ram_addr_t ram_block_add(RAMBlock *new_block, Error **errp)
      * QLIST (which has an RCU-friendly variant) does not have insertion at
      * tail, so save the last element in last_block.
      */
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         last_block = block;
         if (block->max_length < new_block->max_length) {
             break;
         }
     }
     if (block) {
-        QLIST_INSERT_BEFORE(block, new_block, next);
+        QLIST_INSERT_BEFORE_RCU(block, new_block, next);
     } else if (last_block) {
-        QLIST_INSERT_AFTER(last_block, new_block, next);
+        QLIST_INSERT_AFTER_RCU(last_block, new_block, next);
     } else { /* list is empty */
-        QLIST_INSERT_HEAD(&ram_list.blocks, new_block, next);
+        QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next);
     }
     ram_list.mru_block = NULL;
 
+    /* Write list before version */
+    smp_wmb();
     ram_list.version++;
     qemu_mutex_unlock_ramlist();
 
@@ -1552,12 +1566,13 @@ void qemu_ram_free_from_ptr(ram_addr_t addr)
 {
     RAMBlock *block;
 
-    /* This assumes the iothread lock is taken here too.  */
     qemu_mutex_lock_ramlist();
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         if (addr == block->offset) {
-            QLIST_REMOVE(block, next);
+            QLIST_REMOVE_RCU(block, next);
             ram_list.mru_block = NULL;
+            /* Write list before version */
+            smp_wmb();
             ram_list.version++;
             g_free_rcu(block, rcu);
             break;
@@ -1583,17 +1598,17 @@ static void reclaim_ramblock(RAMBlock *block)
     g_free(block);
 }
 
-/* Called with the iothread lock held */
 void qemu_ram_free(ram_addr_t addr)
 {
     RAMBlock *block;
 
-    /* This assumes the iothread lock is taken here too.  */
     qemu_mutex_lock_ramlist();
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         if (addr == block->offset) {
-            QLIST_REMOVE(block, next);
+            QLIST_REMOVE_RCU(block, next);
             ram_list.mru_block = NULL;
+            /* Write list before version */
+            smp_wmb();
             ram_list.version++;
             call_rcu(block, reclaim_ramblock, rcu);
             break;
@@ -1610,7 +1625,7 @@ void qemu_ram_remap(ram_addr_t addr, ram_addr_t length)
     int flags;
     void *area, *vaddr;
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         offset = addr - block->offset;
         if (offset < block->max_length) {
             vaddr = ramblock_ptr(block, offset);
@@ -1657,8 +1672,10 @@ int qemu_get_ram_fd(ram_addr_t addr)
     RAMBlock *block;
     int fd;
 
+    rcu_read_lock();
     block = qemu_get_ram_block(addr);
     fd = block->fd;
+    rcu_read_unlock();
     return fd;
 }
 
@@ -1667,8 +1684,10 @@ void *qemu_get_ram_block_host_ptr(ram_addr_t addr)
     RAMBlock *block;
     void *ptr;
 
+    rcu_read_lock();
     block = qemu_get_ram_block(addr);
     ptr = ramblock_ptr(block, 0);
+    rcu_read_unlock();
     return ptr;
 }
 
@@ -1676,12 +1695,19 @@ void *qemu_get_ram_block_host_ptr(ram_addr_t addr)
  * This should not be used for general purpose DMA.  Use address_space_map
  * or address_space_rw instead. For local memory (e.g. video ram) that the
  * device owns, use memory_region_get_ram_ptr.
+ *
+ * By the time this function returns, the returned pointer is not protected
+ * by RCU anymore.  If the caller is not within an RCU critical section and
+ * does not hold the iothread lock, it must have other means of protecting the
+ * pointer, such as a reference to the region that includes the incoming
+ * ram_addr_t.
  */
 void *qemu_get_ram_ptr(ram_addr_t addr)
 {
     RAMBlock *block;
     void *ptr;
 
+    rcu_read_lock();
     block = qemu_get_ram_block(addr);
 
     if (xen_enabled() && block->host == NULL) {
@@ -1691,19 +1717,26 @@ void *qemu_get_ram_ptr(ram_addr_t addr)
          */
         if (block->offset == 0) {
             ptr = xen_map_cache(addr, 0, 0);
-            goto done;
+            goto unlock;
         }
 
         block->host = xen_map_cache(block->offset, block->max_length, 1);
     }
     ptr = ramblock_ptr(block, addr - block->offset);
 
-done:
+unlock:
+    rcu_read_unlock();
     return ptr;
 }
 
 /* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr
  * but takes a size argument.
+ *
+ * By the time this function returns, the returned pointer is not protected
+ * by RCU anymore.  If the caller is not within an RCU critical section and
+ * does not hold the iothread lock, it must have other means of protecting the
+ * pointer, such as a reference to the region that includes the incoming
+ * ram_addr_t.
  */
 static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
 {
@@ -1715,11 +1748,13 @@ static void *qemu_ram_ptr_length(ram_addr_t addr, hwaddr *size)
         return xen_map_cache(addr, *size, 1);
     } else {
         RAMBlock *block;
-        QLIST_FOREACH(block, &ram_list.blocks, next) {
+        rcu_read_lock();
+        QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
             if (addr - block->offset < block->max_length) {
                 if (addr - block->offset + *size > block->max_length)
                     *size = block->max_length - addr + block->offset;
                 ptr = ramblock_ptr(block, addr - block->offset);
+                rcu_read_unlock();
                 return ptr;
             }
         }
@@ -1745,17 +1780,20 @@ MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
     MemoryRegion *mr;
 
     if (xen_enabled()) {
+        rcu_read_lock();
         *ram_addr = xen_ram_addr_from_mapcache(ptr);
         mr = qemu_get_ram_block(*ram_addr)->mr;
+        rcu_read_unlock();
         return mr;
     }
 
-    block = ram_list.mru_block;
+    rcu_read_lock();
+    block = atomic_rcu_read(&ram_list.mru_block);
     if (block && block->host && host - block->host < block->max_length) {
         goto found;
     }
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         /* This case append when the block is not mapped. */
         if (block->host == NULL) {
             continue;
@@ -1765,11 +1803,13 @@ MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr)
         }
     }
 
+    rcu_read_unlock();
     return NULL;
 
 found:
     *ram_addr = block->offset + (host - block->host);
     mr = block->mr;
+    rcu_read_unlock();
     return mr;
 }
 
@@ -3027,8 +3067,10 @@ void qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque)
 {
     RAMBlock *block;
 
-    QLIST_FOREACH(block, &ram_list.blocks, next) {
+    rcu_read_lock();
+    QLIST_FOREACH_RCU(block, &ram_list.blocks, next) {
         func(block->host, block->offset, block->used_length, opaque);
     }
+    rcu_read_unlock();
 }
 #endif
diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h
index 87b8658..ac06c67 100644
--- a/include/exec/cpu-all.h
+++ b/include/exec/cpu-all.h
@@ -279,9 +279,7 @@ struct RAMBlock {
     uint32_t flags;
     /* Protected by iothread lock.  */
     char idstr[256];
-    /* Reads can take either the iothread or the ramlist lock.
-     * Writes must take both locks.
-     */
+    /* RCU-enabled, writes protected by the ramlist lock */
     QLIST_ENTRY(RAMBlock) next;
     int fd;
 };
@@ -298,7 +296,7 @@ typedef struct RAMList {
     /* Protected by the iothread lock.  */
     unsigned long *dirty_memory[DIRTY_MEMORY_NUM];
     RAMBlock *mru_block;
-    /* Protected by the ramlist lock.  */
+    /* RCU-enabled, writes protected by the ramlist lock. */
     QLIST_HEAD(, RAMBlock) blocks;
     uint32_t version;
 } RAMList;
-- 
2.3.0

Re: [Qemu-devel] [PULL 00/21] SCSI, RCU, KVM changes for 2015-02-16

[Peter Maydell]

On 17 February 2015 at 03:17, Paolo Bonzini <pbonzini@redhat.com> wrote:
> The following changes since commit bc5baffa3554e4c0d20c1dbe879aec931866bd69:
>
>   vfio: Fix debug message compile error (2015-02-10 10:25:44 -0700)
>
> are available in the git repository at:
>
>   git://github.com/bonzini/qemu.git tags/for-upstream
>
> for you to fetch changes up to 0dc3f44aca18b1be8b425f3f4feb4b3e8d68de2e:
>
>   Convert ram_list to RCU (2015-02-16 17:31:55 +0100)
>
> ----------------------------------------------------------------
> - vhost-scsi: add bootindex property
> - RCU: fix MemoryRegion lifetime issues in PCI; document the rules;
> convert of AddressSpaceDispatch and RAMList
> - KVM: add kvm_exit reasons for aarch64
>
> ----------------------------------------------------------------

Applied, thanks.

-- PMM

Re: [Qemu-devel] [PULL 08/21] rcu: run RCU callbacks under the BQL

[Leon Alrae]

Hi,

On 16/02/2015 18:17, Paolo Bonzini wrote:
> This needs to go away sooner or later, but one complication is the
> complex VFIO data structures that are modified in instance_finalize.
> Take a shortcut for now.

It looks like this commit introduced a race condition. QEMU segfaults
very early in one of my tests where I boot Linux, and on one of my
machines it's easier to reproduce than on the others.

(gdb) bt
#0  0x00007ffff7b09fe3 in qemu_cpu_kick_thread (cpu=0x0) at
/slave/smoketest/src/cpus.c:1045
#1  0x00007ffff7b0a269 in qemu_mutex_lock_iothread () at
/slave/smoketest/src/cpus.c:1119
#2  0x00007ffff7ec5c39 in call_rcu_thread (opaque=0x0) at
/slave/smoketest/src/util/rcu.c:241
#3  0x00007ffff38489d1 in start_thread () from /lib64/libpthread.so.0
#4  0x00007ffff359586d in clone () from /lib64/libc.so.6

qemu_mutex_lock_iothread() passes first_cpu to qemu_cpu_kick_thread().
Sometimes this is happening before executing cpu_exec_init(), which I
believe is the reason why qemu_cpu_kick_thread() receives a null pointer.

Regards,
Leon

Re: [Qemu-devel] [PULL 08/21] rcu: run RCU callbacks under the BQL

[Paolo Bonzini]

On 25/02/2015 18:16, Leon Alrae wrote:
> Hi,
> 
> On 16/02/2015 18:17, Paolo Bonzini wrote:
>> This needs to go away sooner or later, but one complication is the
>> complex VFIO data structures that are modified in instance_finalize.
>> Take a shortcut for now.
> 
> It looks like this commit introduced a race condition. QEMU segfaults
> very early in one of my tests where I boot Linux, and on one of my
> machines it's easier to reproduce than on the others.

This should be enough to fix it.  If it works for you I'll send a pull
request as soon as possible.

diff --git a/cpus.c b/cpus.c
index 1cd9867..7bed67b 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1115,7 +1115,7 @@ bool qemu_in_vcpu_thread(void)

 void qemu_mutex_lock_iothread(void)
 {
-    if (!tcg_enabled()) {
+    if (!tcg_enabled() || !first_cpu) {
         qemu_mutex_lock(&qemu_global_mutex);
     } else {
         iothread_requesting_mutex = true;

Paolo

Re: [Qemu-devel] [PULL 08/21] rcu: run RCU callbacks under the BQL

[Leon Alrae]

On 26/02/2015 11:35, Paolo Bonzini wrote:
> 
> 
> On 25/02/2015 18:16, Leon Alrae wrote:
>> Hi,
>>
>> On 16/02/2015 18:17, Paolo Bonzini wrote:
>>> This needs to go away sooner or later, but one complication is the
>>> complex VFIO data structures that are modified in instance_finalize.
>>> Take a shortcut for now.
>>
>> It looks like this commit introduced a race condition. QEMU segfaults
>> very early in one of my tests where I boot Linux, and on one of my
>> machines it's easier to reproduce than on the others.
> 
> This should be enough to fix it.  If it works for you I'll send a pull
> request as soon as possible.
> 
> diff --git a/cpus.c b/cpus.c
> index 1cd9867..7bed67b 100644
> --- a/cpus.c
> +++ b/cpus.c
> @@ -1115,7 +1115,7 @@ bool qemu_in_vcpu_thread(void)
> 
>  void qemu_mutex_lock_iothread(void)
>  {
> -    if (!tcg_enabled()) {
> +    if (!tcg_enabled() || !first_cpu) {
>          qemu_mutex_lock(&qemu_global_mutex);
>      } else {
>          iothread_requesting_mutex = true;

It works for me -- it fixes the crash, thanks.

FYI I hit some other problem where QEMU stops responding (deadlock?)
during a stress test quite randomly. The issue is most likely introduced
by this commit (and above fix doesn't seem to be related). I'm going to
look into that.

Leon

Re: [Qemu-devel] [PULL 08/21] rcu: run RCU callbacks under the BQL

[Paolo Bonzini]

On 26/02/2015 14:40, Leon Alrae wrote:
> FYI I hit some other problem where QEMU stops responding (deadlock?)
> during a stress test quite randomly. The issue is most likely introduced
> by this commit (and above fix doesn't seem to be related). I'm going to
> look into that.

Hmm, I think I see what's going on.  It's related to this commit.

Paolo

Re: [Qemu-devel] [PULL 08/21] rcu: run RCU callbacks under the BQL

[Paolo Bonzini]

On 27/02/2015 15:24, Leon Alrae wrote:
> On 27/02/2015 13:17, Paolo Bonzini wrote:
>> Can you test this patch?  (On top of the previous one).
> 
> With this change the system doesn't boot at all I'm afraid.

Hmm, it works for me and fixes the deadlock with Aurelien's images.

Just to be clear, this is the patch I'm testing on top of
origin/master:

diff --git a/cpus.c b/cpus.c
index 1cd9867..83c078e 100644
--- a/cpus.c
+++ b/cpus.c
@@ -778,7 +778,7 @@ static void qemu_tcg_init_cpu_signals(void)

 static QemuMutex qemu_global_mutex;
 static QemuCond qemu_io_proceeded_cond;
-static bool iothread_requesting_mutex;
+static unsigned iothread_requesting_mutex;

 static QemuThread io_thread;

@@ -1115,15 +1115,15 @@ bool qemu_in_vcpu_thread(void)

 void qemu_mutex_lock_iothread(void)
 {
-    if (!tcg_enabled()) {
+    if (!tcg_enabled() || !first_cpu) {
         qemu_mutex_lock(&qemu_global_mutex);
     } else {
-        iothread_requesting_mutex = true;
+        atomic_inc(&iothread_requesting_mutex);
         if (qemu_mutex_trylock(&qemu_global_mutex)) {
             qemu_cpu_kick_thread(first_cpu);
             qemu_mutex_lock(&qemu_global_mutex);
         }
-        iothread_requesting_mutex = false;
+        atomic_dec(&iothread_requesting_mutex);
         qemu_cond_broadcast(&qemu_io_proceeded_cond);
     }
 }


I couldn't reproduce it with stress, but a reboot loop finally triggered
it on Aurelien's images.

Interestingly, the bug has been there forever but was only triggered in
relatively weird cases such as TCG+migration.  My patch just made it
much more visible.

Paolo

> 
> BTW I managed to reproduce the original problem using Aurel's Debian
> images. Could you try and see if you can reproduce the problem as well?
> 
> Images:
> $ wget https://people.debian.org/~aurel32/qemu/mips/vmlinux-3.2.0-4-4kc-malta
> $ wget https://people.debian.org/~aurel32/qemu/mips/debian_squeeze_mips_standard.qcow2
> 
> Run:
> $ qemu-system-mips -M malta -kernel vmlinux-3.2.0-4-4kc-malta -hda
> debian_squeeze_mips_standard.qcow2 -append "root=/dev/sda1
> console=ttyS0" -snapshot -nographic
> 
> debian-mips login: root
> password: root
> root@debian-mips:~# apt-get install stress
> 
> (and now try to stress a few times, usually after 2 or 3 times QEMU freezes)
> 
> root@debian-mips:~# stress --timeout 15s --cpu 4 --io 2 --vm 2 --vm-bytes 1M