infrastructure for sparse checks for RCU usage
Subsystems (29)
cover infrastructure for sparse checks for RCU usage
Hello!
This patchset pulls Arnd's sparse-checking commits out of the earlier patchbomb (http://lkml.org/lkml/2010/5/4/298):
1. Add an rcu_dereference_check() API in order to continue supporting
the array-index use cases that would otherwise be invalidated by this change.
2. Add an empty __rcu annotation API to reduce inter-commit dependency.
3-5. Updates to vfs, net, and mce to avoid breakage by later commits.
6. Introduce sparse __rcu functionality.
7. Update rculist primitives to avoid false positives.
8-22. Introduce __rcu annotations to cgroups, credentials, keys, nfs,
net, perf_event, notifiers, radix-tree, idr, input, net-netfilter, kvm, and kernel.
23. Introduce __rcu annotations and checks to vhost.
These have been reordered and consolidated to reduce inter-commit dependencies.
Thanx, Paul
b/arch/x86/include/asm/kvm_host.h | 2 b/arch/x86/kernel/cpu/mcheck/mce.c | 2 b/drivers/input/evdev.c | 2 b/drivers/net/bnx2.h | 2 b/drivers/net/bnx2x.h | 2 b/drivers/net/cnic.h | 2 b/drivers/net/macvtap.c | 2 b/drivers/vhost/net.c | 11 - b/drivers/vhost/vhost.c | 14 - b/drivers/vhost/vhost.h | 4 b/include/linux/cgroup.h | 4 b/include/linux/compiler.h | 2 b/include/linux/cred.h | 2 b/include/linux/fdtable.h | 1 b/include/linux/fs.h | 2 b/include/linux/genhd.h | 6 b/include/linux/idr.h | 4 b/include/linux/if_bridge.h | 3 b/include/linux/if_macvlan.h | 2 b/include/linux/igmp.h | 4 b/include/linux/init_task.h | 4 b/include/linux/input.h | 2 b/include/linux/iocontext.h | 2 b/include/linux/key.h | 3 b/include/linux/kvm_host.h | 2 b/include/linux/mm_types.h | 2 b/include/linux/netdevice.h | 12 - b/include/linux/nfs_fs.h | 2 b/include/linux/notifier.h | 10 b/include/linux/perf_event.h | 6 b/include/linux/radix-tree.h | 4 b/include/linux/rculist.h | 53 +++- b/include/linux/rculist_nulls.h | 16 + b/include/linux/rcupdate.h | 33 +++ b/include/linux/sched.h | 2 b/include/linux/srcu.h | 27 ++ b/include/linux/sunrpc/auth_gss.h | 4 b/include/net/dst.h | 2 b/include/net/fib_rules.h | 2 b/include/net/garp.h | 2 b/include/net/inet_sock.h | 2 b/include/net/ip6_tunnel.h | 2 b/include/net/ipip.h | 6 b/include/net/net_namespace.h | 2 b/include/net/netfilter/nf_conntrack.h | 2 b/include/net/netns/xfrm.h | 2 b/include/net/sock.h | 4 b/kernel/cgroup.c | 2 b/kernel/pid.c | 2 b/kernel/rcupdate.c | 6 b/kernel/sched.c | 2 b/lib/Kconfig.debug | 13 + b/lib/radix-tree.c | 2 b/net/802/stp.c | 4 b/net/bridge/br_fdb.c | 2 b/net/bridge/br_private.h | 8 b/net/bridge/netfilter/ebt_redirect.c | 2 b/net/bridge/netfilter/ebt_ulog.c | 4 b/net/bridge/netfilter/ebtables.c | 4 b/net/ipv4/ip_gre.c | 2 b/net/ipv4/ipip.c | 10 b/net/ipv4/netfilter/nf_nat_core.c | 2 b/net/ipv4/protocol.c | 2 b/net/ipv4/route.c | 2 b/net/ipv4/tcp.c | 4 b/net/ipv6/ip6_tunnel.c | 6 b/net/ipv6/protocol.c | 2 b/net/ipv6/sit.c | 10 b/net/mac80211/ieee80211_i.h | 15 - b/net/mac80211/sta_info.h | 4 b/net/netfilter/core.c | 2 b/net/netfilter/nf_conntrack_ecache.c | 4 b/net/netfilter/nf_conntrack_extend.c | 2 b/net/netfilter/nf_conntrack_proto.c | 4 b/net/netfilter/nf_log.c | 2 b/net/netfilter/nf_queue.c | 2 b/net/netfilter/nfnetlink_log.c | 4 b/net/netfilter/nfnetlink_queue.c | 4 b/net/netlabel/netlabel_domainhash.c | 4 b/net/netlabel/netlabel_unlabeled.c | 4 b/net/netlink/af_netlink.c | 2 b/net/phonet/af_phonet.c | 2 b/net/phonet/pn_dev.c | 2 b/net/socket.c | 2 include/linux/compiler.h | 4 include/linux/fdtable.h | 6 include/linux/kvm_host.h | 4 include/linux/rcupdate.h | 352 +++++++++++++++++++-------------- include/linux/sched.h | 6 89 files changed, 498 insertions(+), 307 deletions(-)
RFC PATCH 5/23 mce: convert to rcu_dereference_index_check()
The mce processing applies rcu_dereference_check() to integers used as array indices. This patch therefore moves mce to the new RCU API rcu_dereference_index_check() that avoids the sparse processing that would otherwise result in compiler errors.
--- arch/x86/kernel/cpu/mcheck/mce.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/mcheck/mce.c +1 -1
@@ -50,7 +50,7 @@static DEFINE_MUTEX (mce_read_mutex);#define rcu_dereference_check_mce (p) \ - rcu_dereference_check((p), \ + rcu_dereference_index_check((p), \ rcu_read_lock_sched_held() || \ lockdep_is_held(&mce_read_mutex))
Signature · 1 line
1.7.0.6
RFC PATCH 9/23 credentials: rcu annotation
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/sched.h | 4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-)
include/linux/sched.h +2 -2
@@ -1301,9 +1301,9 @@ struct task_struct {struct list_head cpu_timers[3 ];/* process credentials */ -const struct cred *real_cred;/* objective and real subjective task +const struct cred __rcu *real_cred;/* objective and real subjective task * credentials (COW) */ -const struct cred *cred;/* effective (overridable) subjective task +const struct cred __rcu *cred;/* effective (overridable) subjective task * credentials (COW) */ struct mutex cred_guard_mutex;/* guard against foreign influences on * credential calculations
Signature · 1 line
1.7.0.6
RFC PATCH 8/23 cgroups: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/cgroup.h | 4 ++-- include/linux/sched.h | 2 +- kernel/cgroup.c | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-)
include/linux/cgroup.h +2 -2
@@ -75,7 +75,7 @@ struct cgroup_subsys_state {unsigned long flags;/* ID for this css, if possible */ -struct css_id *id; +struct css_id __rcu *id; };/* bits in struct cgroup_subsys_state flags field */
@@ -205,7 +205,7 @@ struct cgroup {struct list_head children;/* my children */ struct cgroup *parent;/* my parent */ -struct dentry *dentry;/* cgroup fs entry, RCU protected */ +struct dentry __rcu *dentry;/* cgroup fs entry, RCU protected */ /* Private pointers for each registered subsystem */ struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT ];
include/linux/sched.h +1 -1
@@ -1429,7 +1429,7 @@ struct task_struct {#endif #ifdef CONFIG_CGROUPS /* Control Group info protected by css_set_lock */ -struct css_set *cgroups; +struct css_set __rcu *cgroups;/* cg_list protected by css_set_lock and tsk->alloc_lock */ struct list_head cg_list;#endif
kernel/cgroup.c +1 -1
@@ -138,7 +138,7 @@ struct css_id {* is called after synchronize_rcu(). But for safe use, css_is_removed() * css_tryget() should be used for avoiding race. */ -struct cgroup_subsys_state *css; +struct cgroup_subsys_state __rcu *css;/* * ID of this css. */
Signature · 1 line
1.7.0.6
RFC PATCH 23/23 vhost: add __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
Also add rcu_dreference_protected() for code paths where locks are held.
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)
drivers/vhost/net.c +8 -3
@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {
@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);
@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;
drivers/vhost/vhost.c +8 -6
@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }
@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;
@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +
@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);
drivers/vhost/vhost.h +2 -2
@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];
@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;
Signature · 1 line
1.7.0.6
RFC PATCH 3/23 vfs: add fs.h to define struct file
In reply to Paul E. McKenneyThe sparse RCU-pointer annotations require definition of the underlying type of any pointer passed to rcu_dereference() and friends. So fcheck_files() needs "struct file" to be defined, so include fs.h.
--- include/linux/fdtable.h | 1 + 1 files changed, 1 insertions(+), 0 deletions(-)
include/linux/fdtable.h +1 -0
@@ -11,6 +11,7 @@#include <linux/rcupdate.h> #include <linux/types.h> #include <linux/init.h> +#include <linux/fs.h> #include <asm/atomic.h>
Signature · 1 line
1.7.0.6
RFC PATCH 10/23 keys: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/cred.h | 2 +- include/linux/key.h | 3 ++- 2 files changed, 3 insertions(+), 2 deletions(-)
include/linux/cred.h +1 -1
@@ -84,7 +84,7 @@ struct thread_group_cred {atomic_t usage;pid_t tgid;/* thread group process ID */ spinlock_t lock; -struct key *session_keyring;/* keyring inherited over fork */ +struct key __rcu *session_keyring;/* keyring inherited over fork */ struct key *process_keyring;/* keyring private to this process */ struct rcu_head rcu;/* RCU deletion hook */ };
include/linux/key.h +2 -1
@@ -178,8 +178,9 @@ struct key {*/ union {unsigned long value; +void __rcu *rcudata;void *data; -struct keyring_list *subscriptions; +struct keyring_list __rcu *subscriptions; } payload; };
Signature · 1 line
1.7.0.6
RFC PATCH 19/23 kvm: add __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/kvm_host.h | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
include/linux/kvm_host.h +1 -1
@@ -197,7 +197,7 @@ struct kvm {struct mutex irq_lock;#ifdef CONFIG_HAVE_KVM_IRQCHIP -struct kvm_irq_routing_table *irq_routing; +struct kvm_irq_routing_table __rcu *irq_routing;struct hlist_head mask_notifier_list;struct hlist_head irq_ack_notifier_list;#endif
Signature · 1 line
1.7.0.6
RFC PATCH 14/23 notifiers: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/notifier.h | 10 +++++----- 1 files changed, 5 insertions(+), 5 deletions(-)
include/linux/notifier.h +5 -5
@@ -49,28 +49,28 @@struct notifier_block {int (*notifier_call)(struct notifier_block *,unsigned long ,void *); -struct notifier_block *next; +struct notifier_block __rcu *next;int priority; };struct atomic_notifier_head {spinlock_t lock; -struct notifier_block *head; +struct notifier_block __rcu *head; };struct blocking_notifier_head {struct rw_semaphore rwsem; -struct notifier_block *head; +struct notifier_block __rcu *head; };struct raw_notifier_head { -struct notifier_block *head; +struct notifier_block __rcu *head; };struct srcu_notifier_head {struct mutex mutex;struct srcu_struct srcu; -struct notifier_block *head; +struct notifier_block __rcu *head; };#define ATOMIC_INIT_NOTIFIER_HEAD (name) do { \
Signature · 1 line
1.7.0.6
RFC PATCH 12/23 net: __rcu annotations for drivers
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- drivers/net/bnx2.h | 2 +- drivers/net/bnx2x.h | 2 +- drivers/net/cnic.h | 2 +- drivers/net/macvtap.c | 2 +- include/linux/if_macvlan.h | 2 +- 5 files changed, 5 insertions(+), 5 deletions(-)
drivers/net/bnx2.h +1 -1
@@ -6746,7 +6746,7 @@ struct bnx2 {u32 tx_wake_thresh;#ifdef BCM_CNIC -struct cnic_ops *cnic_ops; +struct cnic_ops __rcu *cnic_ops;void *cnic_data;#endif
drivers/net/bnx2x.h +1 -1
@@ -1007,7 +1007,7 @@ struct bnx2x {dma_addr_t timers_mapping;void *qm;dma_addr_t qm_mapping; -struct cnic_ops *cnic_ops; +struct cnic_ops __rcu *cnic_ops;void *cnic_data;u32 cnic_tag;struct cnic_eth_dev cnic_eth_dev;
drivers/net/cnic.h +1 -1
@@ -177,7 +177,7 @@ struct cnic_local {#define ULP_F_INIT 0#define ULP_F_START 1#define ULP_F_CALL_PENDING 2 -struct cnic_ulp_ops *ulp_ops[MAX_CNIC_ULP_TYPE ]; +struct cnic_ulp_ops __rcu *ulp_ops[MAX_CNIC_ULP_TYPE ];/* protected by ulp_lock */ u32 cnic_local_flags;
drivers/net/macvtap.c +1 -1
@@ -37,7 +37,7 @@struct macvtap_queue {struct sock sk;struct socket sock; -struct macvlan_dev *vlan; +struct macvlan_dev __rcu *vlan;struct file *file;unsigned int flags; };
include/linux/if_macvlan.h +1 -1
@@ -47,7 +47,7 @@ struct macvlan_dev {enum macvlan_mode mode;int (*receive)(struct sk_buff *skb);int (*forward)(struct net_device *dev,struct sk_buff *skb); -struct macvtap_queue *tap; +struct macvtap_queue __rcu *tap; }; static inline void macvlan_count_rx(const struct macvlan_dev *vlan,
Signature · 1 line
1.7.0.6
RFC PATCH 22/23 kvm: more __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- arch/x86/include/asm/kvm_host.h | 2 +- include/linux/kvm_host.h | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)
arch/x86/include/asm/kvm_host.h +1 -1
@@ -384,7 +384,7 @@ struct kvm_mem_aliases { };struct kvm_arch { -struct kvm_mem_aliases *aliases; +struct kvm_mem_aliases __rcu *aliases;unsigned int n_free_mmu_pages;unsigned int n_requested_mmu_pages;
include/linux/kvm_host.h +2 -2
@@ -169,7 +169,7 @@ struct kvm {raw_spinlock_t requests_lock;struct mutex slots_lock;struct mm_struct *mm;/* userspace tied to this vm */ -struct kvm_memslots *memslots; +struct kvm_memslots __rcu *memslots;struct srcu_struct srcu;#ifdef CONFIG_KVM_APIC_ARCHITECTURE u32 bsp_vcpu_id;
@@ -179,7 +179,7 @@ struct kvm {atomic_t online_vcpus;struct list_head vm_list;struct mutex lock; -struct kvm_io_bus *buses[KVM_NR_BUSES ]; +struct kvm_io_bus __rcu *buses[KVM_NR_BUSES ]; #ifdef CONFIG_HAVE_KVM_EVENTFD struct { spinlock_t lock;
Signature · 1 line
1.7.0.6
RFC PATCH 17/23 input: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- drivers/input/evdev.c | 2 +- include/linux/input.h | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/input/evdev.c +1 -1
@@ -28,7 +28,7 @@ struct evdev {int minor;struct input_handle handle;wait_queue_head_t wait; -struct evdev_client *grab; +struct evdev_client __rcu *grab;struct list_head client_list;spinlock_t client_lock;/* protects client_list */ struct mutex mutex;
include/linux/input.h +1 -1
@@ -1173,7 +1173,7 @@ struct input_dev {int (*flush)(struct input_dev *dev,struct file *file);int (*event)(struct input_dev *dev,unsigned int type,unsigned int code,int value); -struct input_handle *grab; +struct input_handle __rcu *grab;spinlock_t event_lock;struct mutex mutex;
Signature · 1 line
1.7.0.6
RFC PATCH 7/23 rculist: avoid __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
This avoids warnings from missing __rcu annotations in the rculist implementation, making it possible to use the same lists in both RCU and non-RCU cases.
We can add rculist annotations later, together with lockdep support for rculist, which is missing as well, but that may involve changing all the users.
--- include/linux/rculist.h | 53 ++++++++++++++++++++++++++-------------- include/linux/rculist_nulls.h | 16 ++++++++---- kernel/pid.c | 2 +- 3 files changed, 46 insertions(+), 25 deletions(-)
include/linux/rculist.h +34 -19
@@ -10,6 +10,12 @@#include <linux/rcupdate.h> /* +* return the ->next pointer of a list_head in an rcu safe +* way, we must not access it directly +*/ +#define list_next_rcu (list) (*((struct list_head __rcu **)(&(list)->next))) + +/* * Insert a new entry between two known consecutive entries. * * This is only for internal list manipulation where we know
@@ -20,7 +26,7 @@ static inline void __list_add_rcu(struct list_head *new, { new->next = next; new->prev = prev; -rcu_assign_pointer (prev->next, new); +rcu_assign_pointer (list_next_rcu (prev), new); next->prev = new; }
@@ -138,7 +144,7 @@ static inline void list_replace_rcu(struct list_head *old, { new->next = old->next; new->prev = old->prev; -rcu_assign_pointer (new->prev->next, new); +rcu_assign_pointer (list_next_rcu (new->prev), new); new->next->prev = new; old->prev =LIST_POISON2 ; }
@@ -193,7 +199,7 @@ static inline void list_splice_init_rcu(struct list_head *list,*/ last->next = at; -rcu_assign_pointer (head->next, first); +rcu_assign_pointer (list_next_rcu (head), first); first->prev = head; at->prev = last; }
@@ -208,7 +214,9 @@ static inline void list_splice_init_rcu(struct list_head *list,* primitives such as list_add_rcu() as long as it's guarded by rcu_read_lock(). */ #define list_entry_rcu (ptr, type, member) \ - container_of(rcu_dereference_raw(ptr), type, member) + ({typeof (*ptr) __rcu *__ptr = (typeof (*ptr) __rcu __force *)ptr; \ + container_of((typeof(ptr))rcu_dereference_raw(__ptr), type, member); \ + })/** * list_first_entry_rcu - get the first element from a list
@@ -225,9 +233,9 @@ static inline void list_splice_init_rcu(struct list_head *list, list_entry_rcu((ptr)->next, type, member)#define __list_for_each_rcu (pos, head) \ - for (pos = rcu_dereference_raw((head)->next); \ + for (pos = rcu_dereference_raw(list_next_rcu(head)); \ pos != (head); \ - pos = rcu_dereference_raw(pos->next)) + pos = rcu_dereference_raw(list_next_rcu((pos)))/** * list_for_each_entry_rcu - iterate over rcu list of given type
@@ -257,9 +265,9 @@ static inline void list_splice_init_rcu(struct list_head *list,* as long as the traversal is guarded by rcu_read_lock(). */ #define list_for_each_continue_rcu (pos, head) \ - for ((pos) = rcu_dereference_raw((pos)->next); \ + for ((pos) = rcu_dereference_raw(list_next_rcu(pos)); \ prefetch((pos)->next), (pos) != (head); \ - (pos) = rcu_dereference_raw((pos)->next)) + (pos) = rcu_dereference_raw(list_next_rcu(pos)))/** * list_for_each_entry_continue_rcu - continue iteration over list of given type
@@ -314,12 +322,19 @@ static inline void hlist_replace_rcu(struct hlist_node *old, new->next = next; new->pprev = old->pprev; -rcu_assign_pointer (*new->pprev, new); +rcu_assign_pointer (*(struct hlist_node __rcu **)new->pprev, new);if (next) new->next->pprev = &new->next; old->pprev =LIST_POISON2 ; } +/* +* return the first or the next element in an RCU protected hlist +*/ +#define hlist_first_rcu (head) (*((struct hlist_node __rcu **)(&(head)->first))) +#define hlist_next_rcu (node) (*((struct hlist_node __rcu **)(&(node)->next))) +#define hlist_pprev_rcu (node) (*((struct hlist_node __rcu **)((node)->pprev))) +/** * hlist_add_head_rcu * @n: the element to add to the hash list.
@@ -346,7 +361,7 @@ static inline void hlist_add_head_rcu(struct hlist_node *n, n->next = first; n->pprev = &h->first; -rcu_assign_pointer (h->first, n); +rcu_assign_pointer (hlist_first_rcu (h), n);if (first) first->pprev = &n->next; }
@@ -374,7 +389,7 @@ static inline void hlist_add_before_rcu(struct hlist_node *n, { n->pprev = next->pprev; n->next = next; -rcu_assign_pointer (*(n->pprev), n); +rcu_assign_pointer (hlist_pprev_rcu (n), n); next->pprev = &n->next; }
@@ -401,15 +416,15 @@ static inline void hlist_add_after_rcu(struct hlist_node *prev, { n->next = prev->next; n->pprev = &prev->next; -rcu_assign_pointer (prev->next, n); +rcu_assign_pointer (hlist_next_rcu (prev), n);if (n->next) n->next->pprev = &n->next; } -#define __hlist_for_each_rcu (pos, head) \ - for (pos = rcu_dereference((head)->first); \ - pos && ({ prefetch(pos->next); 1; }); \ - pos = rcu_dereference(pos->next)) +#define __hlist_for_each_rcu (pos, head) \ + for (pos = rcu_dereference(hlist_first_rcu(head)); \ + pos && ({ prefetch(pos->next); 1; }); \ + pos = rcu_dereference(hlist_next_rcu(pos)))/** * hlist_for_each_entry_rcu - iterate over rcu list of given type
@@ -422,11 +437,11 @@ static inline void hlist_add_after_rcu(struct hlist_node *prev,* the _rcu list-mutation primitives such as hlist_add_head_rcu() * as long as the traversal is guarded by rcu_read_lock(). */ -#define hlist_for_each_entry_rcu (tpos, pos, head, member) \ - for (pos = rcu_dereference_raw((head)->first); \ +#define hlist_for_each_entry_rcu (tpos, pos, head, member) \ + for (pos = rcu_dereference_raw(hlist_first_rcu(head)); \ pos && ({ prefetch(pos->next); 1; }) && \ ({ tpos = hlist_entry(pos, typeof(*tpos), member); 1; }); \ - pos = rcu_dereference_raw(pos->next)) + pos = rcu_dereference_raw(hlist_next_rcu(pos)))#endif /* __KERNEL__ */ #endif
include/linux/rculist_nulls.h +11 -5
@@ -37,6 +37,12 @@ static inline void hlist_nulls_del_init_rcu(struct hlist_nulls_node *n) } } +#define hlist_nulls_first_rcu (head) \ + (*((struct hlist_nulls_node __rcu __force **)&(head)->first)) + +#define hlist_nulls_next_rcu (node) \ + (*((struct hlist_nulls_node __rcu __force **)&(node)->next)) +/** * hlist_nulls_del_rcu - deletes entry from hash list without re-initialization * @n: the element to delete from the hash list.
@@ -88,7 +94,7 @@ static inline void hlist_nulls_add_head_rcu(struct hlist_nulls_node *n, n->next = first; n->pprev = &h->first; -rcu_assign_pointer (h->first, n); +rcu_assign_pointer (hlist_nulls_first_rcu (h), n);if (!is_a_nulls (first)) first->pprev = &n->next; }
@@ -100,11 +106,11 @@ static inline void hlist_nulls_add_head_rcu(struct hlist_nulls_node *n,* @member: the name of the hlist_nulls_node within the struct. * */ -#define hlist_nulls_for_each_entry_rcu (tpos, pos, head, member) \ - for (pos = rcu_dereference_raw((head)->first); \ - (!is_a_nulls(pos)) && \ +#define hlist_nulls_for_each_entry_rcu (tpos, pos, head, member) \ + for (pos = rcu_dereference_raw(hlist_nulls_first_rcu(head)); \ + (!is_a_nulls(pos)) && \ ({ tpos = hlist_nulls_entry(pos, typeof(*tpos), member); 1; }); \ - pos = rcu_dereference_raw(pos->next)) + pos = rcu_dereference_raw(hlist_nulls_next_rcu(pos)))#endif #endif
kernel/pid.c +1 -1
@@ -367,7 +367,7 @@ struct task_struct *pid_task(struct pid *pid, enum pid_type type)struct task_struct *result =NULL ;if (pid) {struct hlist_node *first; - first =rcu_dereference_check (pid->tasks[type].first, + first =rcu_dereference_check (hlist_first_rcu (&pid->tasks[type]),rcu_read_lock_held () ||lockdep_tasklist_lock_is_held ());if (first)
Signature · 1 line
1.7.0.6
RFC PATCH 20/23 kernel: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
This adds annotations for RCU operations in core kernel components
--- include/linux/fdtable.h | 6 +++--- include/linux/fs.h | 2 +- include/linux/genhd.h | 6 +++--- include/linux/init_task.h | 4 ++-- include/linux/iocontext.h | 2 +- include/linux/mm_types.h | 2 +- 6 files changed, 11 insertions(+), 11 deletions(-)
include/linux/fdtable.h +3 -3
@@ -31,7 +31,7 @@ struct embedded_fd_set {struct fdtable {unsigned int max_fds; -struct file ** fd;/* current fd array */ +struct file __rcu **fd;/* current fd array */ fd_set *close_on_exec;fd_set *open_fds;struct rcu_head rcu;
@@ -46,7 +46,7 @@ struct files_struct {* read mostly part */ atomic_t count; -struct fdtable *fdt; +struct fdtable __rcu *fdt;struct fdtable fdtab;/* * written part on a separate cache line in SMP
@@ -55,7 +55,7 @@ struct files_struct {int next_fd;struct embedded_fd_set close_on_exec_init;struct embedded_fd_set open_fds_init; -struct file * fd_array[NR_OPEN_DEFAULT ]; +struct file __rcu * fd_array[NR_OPEN_DEFAULT ]; };#define rcu_dereference_check_fdtable (files, fdtfd) \
include/linux/fs.h +1 -1
@@ -1382,7 +1382,7 @@ struct super_block {* Saved mount options for lazy filesystems using * generic_show_options() */ -char *s_options; +char __rcu *s_options; };extern struct timespec current_fs_time (struct super_block *sb);
include/linux/genhd.h +3 -3
@@ -129,8 +129,8 @@ struct blk_scsi_cmd_filter {struct disk_part_tbl {struct rcu_head rcu_head;int len; -struct hd_struct *last_lookup; -struct hd_struct *part[]; +struct hd_struct __rcu *last_lookup; +struct hd_struct __rcu *part[]; };struct gendisk {
@@ -149,7 +149,7 @@ struct gendisk {* non-critical accesses use RCU. Always access through * helpers. */ -struct disk_part_tbl *part_tbl; +struct disk_part_tbl __rcu *part_tbl;struct hd_struct part0;const struct block_device_operations *fops;
include/linux/init_task.h +2 -2
@@ -137,8 +137,8 @@ extern struct cred init_cred; .children = LIST_HEAD_INIT(tsk.children), \ .sibling = LIST_HEAD_INIT(tsk.sibling), \ .group_leader = &tsk, \ - .real_cred = &init_cred, \ - .cred = &init_cred, \ + RCU_INIT_POINTER(.real_cred, &init_cred), \ + RCU_INIT_POINTER(.cred, &init_cred), \ .cred_guard_mutex = \__MUTEX_INITIALIZER (tsk.cred_guard_mutex), \ .comm = "swapper", \
include/linux/iocontext.h +1 -1
@@ -54,7 +54,7 @@ struct io_context {struct radix_tree_root radix_root;struct hlist_head cic_list; -void *ioc_data; +void __rcu *ioc_data; };static inline struct io_context *ioc_task_link (struct io_context *ioc)
include/linux/mm_types.h +1 -1
@@ -299,7 +299,7 @@ struct mm_struct {* new_owner->mm == mm * new_owner->alloc_lock is held */ -struct task_struct *owner; +struct task_struct __rcu *owner;#endif #ifdef CONFIG_PROC_FS
Signature · 1 line
1.7.0.6
RFC PATCH 21/23 net: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/igmp.h | 4 ++-- include/linux/netdevice.h | 12 ++++++------ include/net/dst.h | 2 +- include/net/fib_rules.h | 2 +- include/net/garp.h | 2 +- include/net/inet_sock.h | 2 +- include/net/ip6_tunnel.h | 2 +- include/net/ipip.h | 6 +++--- include/net/net_namespace.h | 2 +- include/net/netns/xfrm.h | 2 +- include/net/sock.h | 4 ++-- kernel/sched.c | 2 +- net/802/stp.c | 4 ++-- net/ipv4/ip_gre.c | 2 +- net/ipv4/ipip.c | 10 +++++----- net/ipv4/protocol.c | 2 +- net/ipv4/route.c | 2 +- net/ipv4/tcp.c | 4 ++-- net/ipv6/ip6_tunnel.c | 6 +++--- net/ipv6/protocol.c | 2 +- net/ipv6/sit.c | 10 +++++----- net/mac80211/ieee80211_i.h | 15 ++++++++------- net/mac80211/sta_info.h | 4 ++-- net/netlabel/netlabel_domainhash.c | 4 ++-- net/netlabel/netlabel_unlabeled.c | 4 ++-- net/netlink/af_netlink.c | 2 +- net/phonet/af_phonet.c | 2 +- net/phonet/pn_dev.c | 2 +- net/socket.c | 2 +- 29 files changed, 60 insertions(+), 59 deletions(-)
include/linux/igmp.h +2 -2
@@ -167,10 +167,10 @@ struct ip_sf_socklist {*/ struct ip_mc_socklist { -struct ip_mc_socklist *next; +struct ip_mc_socklist __rcu *next;struct ip_mreqn multi;unsigned int sfmode;/* MCAST_{INCLUDE,EXCLUDE} */ -struct ip_sf_socklist *sflist; +struct ip_sf_socklist __rcu *sflist;struct rcu_head rcu; };
include/linux/netdevice.h +6 -6
@@ -855,10 +855,10 @@ struct net_device {#ifdef CONFIG_NET_DSA void *dsa_ptr;/* dsa specific data */ #endif -void *atalk_ptr;/* AppleTalk link */ -void *ip_ptr;/* IPv4 specific data */ +void *atalk_ptr;/* AppleTalk link */ +void __rcu *ip_ptr;/* IPv4 specific data */ void *dn_ptr;/* DECnet specific data */ -void *ip6_ptr;/* IPv6 specific data */ +void __rcu *ip6_ptr;/* IPv6 specific data */ void *ec_ptr;/* Econet specific data */ void *ax25_ptr;/* AX.25 specific data */ struct wireless_dev *ieee80211_ptr;/* IEEE 802.11 specific data,
@@ -947,11 +947,11 @@ struct net_device {void *ml_priv;/* bridge stuff */ -struct net_bridge_port *br_port; +void __rcu *br_port;/* macvlan */ -struct macvlan_port *macvlan_port; +struct macvlan_port __rcu *macvlan_port;/* GARP */ -struct garp_port *garp_port; +struct garp_port __rcu *garp_port;/* class/net/name entry */ struct device dev;
include/net/dst.h +1 -1
@@ -94,7 +94,7 @@ struct dst_entry {unsigned long lastuse;union {struct dst_entry *next; -struct rtable *rt_next; +struct rtable __rcu *rt_next;struct rt6_info *rt6_next;struct dn_route *dn_next; };
include/net/fib_rules.h +1 -1
@@ -20,7 +20,7 @@ struct fib_rule {u32 table;u8 action;u32 target; -struct fib_rule * ctarget; +struct fib_rule __rcu *ctarget;char iifname[IFNAMSIZ ];char oifname[IFNAMSIZ ];struct rcu_head rcu;
include/net/garp.h +1 -1
@@ -107,7 +107,7 @@ struct garp_applicant { };struct garp_port { -struct garp_applicant *applicants[GARP_APPLICATION_MAX +1 ]; +struct garp_applicant __rcu *applicants[GARP_APPLICATION_MAX +1 ]; };extern int garp_register_application (struct garp_application *app);
include/net/inet_sock.h +1 -1
@@ -139,7 +139,7 @@ struct inet_sock { mc_all:1 ;int mc_index;__be32 mc_addr; -struct ip_mc_socklist *mc_list; +struct ip_mc_socklist __rcu *mc_list;struct {unsigned int flags;unsigned int fragsize;
include/net/ip6_tunnel.h +1 -1
@@ -13,7 +13,7 @@/* IPv6 tunnel */ struct ip6_tnl { -struct ip6_tnl *next;/* next tunnel in list */ +struct ip6_tnl __rcu *next;/* next tunnel in list */ struct net_device *dev;/* virtual device associated with tunnel */ struct ip6_tnl_parm parms;/* tunnel configuration parameters */ struct flowi fl;/* flowi template for xmit */
include/net/ipip.h +3 -3
@@ -16,7 +16,7 @@ struct ip_tunnel_6rd_parm { };struct ip_tunnel { -struct ip_tunnel *next; +struct ip_tunnel __rcu *next;struct net_device *dev;int err_count;/* Number of arrived ICMP errors */
@@ -34,12 +34,12 @@ struct ip_tunnel {#ifdef CONFIG_IPV6_SIT_6RD struct ip_tunnel_6rd_parm ip6rd;#endif -struct ip_tunnel_prl_entry *prl;/* potential router list */ +struct ip_tunnel_prl_entry __rcu *prl;/* potential router list */ unsigned int prl_count;/* # of entries in PRL */ };struct ip_tunnel_prl_entry { -struct ip_tunnel_prl_entry *next; +struct ip_tunnel_prl_entry __rcu *next;__be32 addr;u16 flags;struct rcu_head rcu_head;
include/net/net_namespace.h +1 -1
@@ -90,7 +90,7 @@ struct net {#ifdef CONFIG_WEXT_CORE struct sk_buff_head wext_nlevents;#endif -struct net_generic *gen; +struct net_generic __rcu *gen; };
include/net/netns/xfrm.h +1 -1
@@ -48,7 +48,7 @@ struct netns_xfrm {struct dst_ops xfrm6_dst_ops;#endif -struct sock *nlsk; +struct sock __rcu *nlsk;struct sock *nlsk_stash;u32 sysctl_aevent_etime;
include/net/sock.h +2 -2
@@ -148,7 +148,7 @@ struct sock_common { };struct proto *skc_prot;#ifdef CONFIG_NET_NS -struct net *skc_net; +struct net __rcu *skc_net;#endif };
@@ -293,7 +293,7 @@ struct sock {struct ucred sk_peercred;long sk_rcvtimeo;long sk_sndtimeo; -struct sk_filter *sk_filter; +struct sk_filter __rcu *sk_filter;void *sk_protinfo;struct timer_list sk_timer;ktime_t sk_stamp;
kernel/sched.c +1 -1
@@ -539,7 +539,7 @@ struct rq {#ifdef CONFIG_SMP struct root_domain *rd; -struct sched_domain *sd; +struct sched_domain __rcu *sd;unsigned char idle_at_tick;/* For active balancing */
net/802/stp.c +2 -2
@@ -21,8 +21,8 @@#define GARP_ADDR_MAX 0x2F#define GARP_ADDR_RANGE (GARP_ADDR_MAX - GARP_ADDR_MIN) -static const struct stp_proto *garp_protos[GARP_ADDR_RANGE +1 ]__read_mostly ; -static const struct stp_proto *stp_proto__read_mostly ; +static const struct stp_proto __rcu *garp_protos[GARP_ADDR_RANGE +1 ]__read_mostly ; +static const struct stp_proto __rcu *stp_proto__read_mostly ;static struct llc_sap *sap__read_mostly ;static unsigned int sap_registered;
net/ipv4/ip_gre.c +1 -1
@@ -128,7 +128,7 @@ static int ipgre_tunnel_bind_dev(struct net_device *dev);static int ipgre_net_id__read_mostly ;struct ipgre_net { -struct ip_tunnel *tunnels[4 ][HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels[4 ][HASH_SIZE ];struct net_device *fb_tunnel_dev; };
net/ipv4/ipip.c +5 -5
@@ -122,11 +122,11 @@static int ipip_net_id__read_mostly ;struct ipip_net { -struct ip_tunnel *tunnels_r_l[HASH_SIZE ]; -struct ip_tunnel *tunnels_r[HASH_SIZE ]; -struct ip_tunnel *tunnels_l[HASH_SIZE ]; -struct ip_tunnel *tunnels_wc[1 ]; -struct ip_tunnel **tunnels[4 ]; +struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels_r[HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels_l[HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels_wc[1 ]; +struct ip_tunnel __rcu **tunnels[4 ];struct net_device *fb_tunnel_dev; };
net/ipv4/protocol.c +1 -1
@@ -28,7 +28,7 @@#include <linux/spinlock.h> #include <net/protocol.h> -const struct net_protocol *inet_protos[MAX_INET_PROTOS ]____cacheline_aligned_in_smp ; +const struct net_protocol __rcu *inet_protos[MAX_INET_PROTOS ]____cacheline_aligned_in_smp ;static DEFINE_SPINLOCK (inet_proto_lock);/*
net/ipv4/route.c +1 -1
@@ -200,7 +200,7 @@ const __u8 ip_tos2prio[16] = {*/ struct rt_hash_bucket { -struct rtable *chain; +struct rtable __rcu *chain; }; #if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK) || \
net/ipv4/tcp.c +2 -2
@@ -3057,9 +3057,9 @@ static struct tcp_cookie_secret tcp_secret_one;static struct tcp_cookie_secret tcp_secret_two;/* Essentially a circular list, without dynamic allocation. */ -static struct tcp_cookie_secret *tcp_secret_generating; +static struct tcp_cookie_secret __rcu *tcp_secret_generating;static struct tcp_cookie_secret *tcp_secret_primary; -static struct tcp_cookie_secret *tcp_secret_retiring; +static struct tcp_cookie_secret __rcu *tcp_secret_retiring;static struct tcp_cookie_secret *tcp_secret_secondary;static DEFINE_SPINLOCK (tcp_secret_locker);
net/ipv6/ip6_tunnel.c +3 -3
@@ -83,9 +83,9 @@ struct ip6_tnl_net {/* the IPv6 tunnel fallback device */ struct net_device *fb_tnl_dev;/* lists for storing tunnels in use */ -struct ip6_tnl *tnls_r_l[HASH_SIZE ]; -struct ip6_tnl *tnls_wc[1 ]; -struct ip6_tnl **tnls[2 ]; +struct ip6_tnl __rcu *tnls_r_l[HASH_SIZE ]; +struct ip6_tnl __rcu *tnls_wc[1 ]; +struct ip6_tnl __rcu **tnls[2 ]; };/*
net/ipv6/protocol.c +1 -1
@@ -25,7 +25,7 @@#include <linux/spinlock.h> #include <net/protocol.h> -const struct inet6_protocol *inet6_protos[MAX_INET_PROTOS ]; +const struct inet6_protocol __rcu *inet6_protos[MAX_INET_PROTOS ];static DEFINE_SPINLOCK (inet6_proto_lock);
net/ipv6/sit.c +5 -5
@@ -68,11 +68,11 @@ static void ipip6_tunnel_setup(struct net_device *dev);static int sit_net_id__read_mostly ;struct sit_net { -struct ip_tunnel *tunnels_r_l[HASH_SIZE ]; -struct ip_tunnel *tunnels_r[HASH_SIZE ]; -struct ip_tunnel *tunnels_l[HASH_SIZE ]; -struct ip_tunnel *tunnels_wc[1 ]; -struct ip_tunnel **tunnels[4 ]; +struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels_r[HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels_l[HASH_SIZE ]; +struct ip_tunnel __rcu *tunnels_wc[1 ]; +struct ip_tunnel __rcu **tunnels[4 ];struct net_device *fb_tunnel_dev; };
net/mac80211/ieee80211_i.h +8 -7
@@ -192,7 +192,7 @@ struct beacon_data { };struct ieee80211_if_ap { -struct beacon_data *beacon; +struct beacon_data __rcu *beacon;struct list_head vlans;
@@ -214,7 +214,7 @@ struct ieee80211_if_vlan {struct list_head list;/* used for all tx if the VLAN is configured to 4-addr mode */ -struct sta_info *sta; +struct sta_info __rcu *sta; };struct mesh_stats {
@@ -388,7 +388,8 @@ struct ieee80211_if_ibss {unsigned long ibss_join_req;/* probe response/beacon for IBSS */ -struct sk_buff *presp, *skb; +struct sk_buff __rcu *presp; +struct sk_buff *skb;enum {IEEE80211_IBSS_MLME_SEARCH ,
@@ -492,9 +493,9 @@ struct ieee80211_sub_if_data {#define NUM_DEFAULT_KEYS 4#define NUM_DEFAULT_MGMT_KEYS 2 -struct ieee80211_key *keys[NUM_DEFAULT_KEYS +NUM_DEFAULT_MGMT_KEYS ]; -struct ieee80211_key *default_key; -struct ieee80211_key *default_mgmt_key; +struct ieee80211_key __rcu *keys[NUM_DEFAULT_KEYS +NUM_DEFAULT_MGMT_KEYS ]; +struct ieee80211_key __rcu *default_key; +struct ieee80211_key __rcu *default_mgmt_key;u16 sequence_number;
@@ -698,7 +699,7 @@ struct ieee80211_local {spinlock_t sta_lock;unsigned long num_sta;struct list_head sta_list, sta_pending_list; -struct sta_info *sta_hash[STA_HASH_SIZE ]; +struct sta_info __rcu *sta_hash[STA_HASH_SIZE ];struct timer_list sta_cleanup;struct work_struct sta_finish_work;int sta_generation;
net/mac80211/sta_info.h +2 -2
@@ -229,10 +229,10 @@ struct sta_ampdu_mlme {struct sta_info {/* General information, mostly static */ struct list_head list; -struct sta_info *hnext; +struct sta_info __rcu *hnext;struct ieee80211_local *local;struct ieee80211_sub_if_data *sdata; -struct ieee80211_key *key; +struct ieee80211_key __rcu *key;struct rate_control_ref *rate_ctrl;void *rate_ctrl_priv;spinlock_t lock;
net/netlabel/netlabel_domainhash.c +2 -2
@@ -57,8 +57,8 @@ static DEFINE_SPINLOCK(netlbl_domhsh_lock);#define netlbl_domhsh_rcu_deref (p) \ rcu_dereference_check(p, rcu_read_lock_held() || \ lockdep_is_held(&netlbl_domhsh_lock)) -static struct netlbl_domhsh_tbl *netlbl_domhsh =NULL ; -static struct netlbl_dom_map *netlbl_domhsh_def =NULL ; +static struct netlbl_domhsh_tbl __rcu *netlbl_domhsh =NULL ; +static struct netlbl_dom_map __rcu *netlbl_domhsh_def =NULL ;/* * Domain Hash Table Helper Functions
net/netlabel/netlabel_unlabeled.c +2 -2
@@ -118,8 +118,8 @@ static DEFINE_SPINLOCK(netlbl_unlhsh_lock);#define netlbl_unlhsh_rcu_deref (p) \ rcu_dereference_check(p, rcu_read_lock_held() || \ lockdep_is_held(&netlbl_unlhsh_lock)) -static struct netlbl_unlhsh_tbl *netlbl_unlhsh =NULL ; -static struct netlbl_unlhsh_iface *netlbl_unlhsh_def =NULL ; +static struct netlbl_unlhsh_tbl __rcu *netlbl_unlhsh =NULL ; +static struct netlbl_unlhsh_iface __rcu *netlbl_unlhsh_def =NULL ;/* Accept unlabeled packets flag */ static u8 netlabel_unlabel_acceptflg =0 ;
net/netlink/af_netlink.c +1 -1
@@ -119,7 +119,7 @@ struct nl_pid_hash {struct netlink_table {struct nl_pid_hash hash;struct hlist_head mc_list; -unsigned long *listeners; +unsigned long __rcu *listeners;unsigned int nl_nonroot;unsigned int groups;struct mutex *cb_mutex;
net/phonet/af_phonet.c +1 -1
@@ -35,7 +35,7 @@#include <net/phonet/pn_dev.h> /* Transport protocol registration */ -static struct phonet_protocol *proto_tab[PHONET_NPROTO ]__read_mostly ; +static struct phonet_protocol __rcu *proto_tab[PHONET_NPROTO ]__read_mostly ;static struct phonet_protocol *phonet_proto_get (int protocol) {
net/phonet/pn_dev.c +1 -1
@@ -36,7 +36,7 @@struct phonet_routes {struct mutex lock; -struct net_device *table[64 ]; +struct net_device __rcu *table[64 ]; };struct phonet_net {
net/socket.c +1 -1
@@ -155,7 +155,7 @@ static const struct file_operations socket_file_ops = {*/ static DEFINE_SPINLOCK (net_family_lock); -static const struct net_proto_family *net_families[NPROTO ]__read_mostly ; +static const struct net_proto_family __rcu *net_families[NPROTO ]__read_mostly ;/* * Statistics counters of the socket lists
Signature · 1 line
1.7.0.6
RFC PATCH 18/23 net/netfilter: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/net/netfilter/nf_conntrack.h | 2 +- net/ipv4/netfilter/nf_nat_core.c | 2 +- net/netfilter/core.c | 2 +- net/netfilter/nf_conntrack_ecache.c | 4 ++-- net/netfilter/nf_conntrack_extend.c | 2 +- net/netfilter/nf_conntrack_proto.c | 4 ++-- net/netfilter/nf_log.c | 2 +- net/netfilter/nf_queue.c | 2 +- 8 files changed, 10 insertions(+), 10 deletions(-)
include/net/netfilter/nf_conntrack.h +1 -1
@@ -75,7 +75,7 @@ struct nf_conntrack_helper;/* nf_conn feature for connections that have a helper */ struct nf_conn_help {/* Helper. if any */ -struct nf_conntrack_helper *helper; +struct nf_conntrack_helper __rcu *helper;union nf_conntrack_help help;
net/ipv4/netfilter/nf_nat_core.c +1 -1
@@ -38,7 +38,7 @@ static DEFINE_SPINLOCK(nf_nat_lock);static struct nf_conntrack_l3proto *l3proto__read_mostly ;#define MAX_IP_NAT_PROTO 256 -static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO ] +static const struct nf_nat_protocol __rcu *nf_nat_protos[MAX_IP_NAT_PROTO ]__read_mostly ; static inline const struct nf_nat_protocol *
net/netfilter/core.c +1 -1
@@ -27,7 +27,7 @@static DEFINE_MUTEX (afinfo_mutex); -const struct nf_afinfo *nf_afinfo[NFPROTO_NUMPROTO ]__read_mostly ; +const struct nf_afinfo __rcu *nf_afinfo[NFPROTO_NUMPROTO ]__read_mostly ;EXPORT_SYMBOL (nf_afinfo);int nf_register_afinfo (const struct nf_afinfo *afinfo)
net/netfilter/nf_conntrack_ecache.c +2 -2
@@ -26,10 +26,10 @@static DEFINE_MUTEX (nf_ct_ecache_mutex); -struct nf_ct_event_notifier *nf_conntrack_event_cb__read_mostly ; +struct nf_ct_event_notifier __rcu *nf_conntrack_event_cb__read_mostly ;EXPORT_SYMBOL_GPL (nf_conntrack_event_cb); -struct nf_exp_event_notifier *nf_expect_event_cb__read_mostly ; +struct nf_exp_event_notifier __rcu *nf_expect_event_cb__read_mostly ;EXPORT_SYMBOL_GPL (nf_expect_event_cb);/* deliver cached events and clear cache entry - must be called with locally
net/netfilter/nf_conntrack_extend.c +1 -1
@@ -16,7 +16,7 @@#include <linux/skbuff.h> #include <net/netfilter/nf_conntrack_extend.h> -static struct nf_ct_ext_type *nf_ct_ext_types[NF_CT_EXT_NUM ]; +static struct nf_ct_ext_type __rcu *nf_ct_ext_types[NF_CT_EXT_NUM ];static DEFINE_MUTEX (nf_ct_ext_type_mutex);void __nf_ct_ext_destroy (struct nf_conn *ct)
net/netfilter/nf_conntrack_proto.c +2 -2
@@ -30,8 +30,8 @@#include <net/netfilter/nf_conntrack_l4proto.h> #include <net/netfilter/nf_conntrack_core.h> -static struct nf_conntrack_l4proto **nf_ct_protos[PF_MAX ]__read_mostly ; -struct nf_conntrack_l3proto *nf_ct_l3protos[AF_MAX ]__read_mostly ; +static struct nf_conntrack_l4proto __rcu **nf_ct_protos[PF_MAX ]__read_mostly ; +struct nf_conntrack_l3proto __rcu *nf_ct_l3protos[AF_MAX ]__read_mostly ;EXPORT_SYMBOL_GPL (nf_ct_l3protos);static DEFINE_MUTEX (nf_ct_proto_mutex);
net/netfilter/nf_log.c +1 -1
@@ -16,7 +16,7 @@#define NF_LOG_PREFIXLEN 128#define NFLOGGER_NAME_LEN 64 -static const struct nf_logger *nf_loggers[NFPROTO_NUMPROTO ]__read_mostly ; +static const struct nf_logger __rcu *nf_loggers[NFPROTO_NUMPROTO ]__read_mostly ;static struct list_head nf_loggers_l[NFPROTO_NUMPROTO ]__read_mostly ;static DEFINE_MUTEX (nf_log_mutex);
net/netfilter/nf_queue.c +1 -1
@@ -17,7 +17,7 @@* long term mutex. The handler must provide an an outfn() to accept packets * for queueing and must reinject all packets it receives, no matter what. */ -static const struct nf_queue_handler *queue_handler[NFPROTO_NUMPROTO ]__read_mostly ; +static const struct nf_queue_handler __rcu *queue_handler[NFPROTO_NUMPROTO ]__read_mostly ;static DEFINE_MUTEX (queue_handler_mutex);
Signature · 1 line
1.7.0.6
RFC PATCH 16/23 idr: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/idr.h | 4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-)
include/linux/idr.h +2 -2
@@ -50,14 +50,14 @@struct idr_layer {unsigned long bitmap;/* A zero bit means "space here" */ -struct idr_layer *ary[1 <<IDR_BITS ]; +struct idr_layer __rcu *ary[1 <<IDR_BITS ];int count;/* When zero, we can release it */ int layer;/* distance from leaf */ struct rcu_head rcu_head; };struct idr { -struct idr_layer *top; +struct idr_layer __rcu *top;struct idr_layer *id_free;int layers;/* only valid without concurrent changes */ int id_free_cnt;
Signature · 1 line
1.7.0.6
RFC PATCH 13/23 perf_event: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/perf_event.h | 6 +++--- include/linux/sched.h | 2 +- 2 files changed, 4 insertions(+), 4 deletions(-)
include/linux/perf_event.h +3 -3
@@ -582,7 +582,7 @@ struct perf_event {int nr_siblings;int group_flags;struct perf_event *group_leader; -struct perf_event *output; +struct perf_event __rcu *output;const struct pmu *pmu;enum perf_event_active_state state;
@@ -643,7 +643,7 @@ struct perf_event {/* mmap bits */ struct mutex mmap_mutex;atomic_t mmap_count; -struct perf_mmap_data *data; +struct perf_mmap_data __rcu *data;/* poll related */ wait_queue_head_t waitq;
@@ -710,7 +710,7 @@ struct perf_event_context {* These fields let us detect when two contexts have both * been cloned (inherited) from a common ancestor. */ -struct perf_event_context *parent_ctx; +struct perf_event_context __rcu *parent_ctx;u64 parent_gen;u64 generation;int pin_count;
include/linux/sched.h +1 -1
@@ -1442,7 +1442,7 @@ struct task_struct {struct futex_pi_state *pi_state_cache;#endif #ifdef CONFIG_PERF_EVENTS -struct perf_event_context *perf_event_ctxp; +struct perf_event_context __rcu *perf_event_ctxp;struct mutex perf_event_mutex;struct list_head perf_event_list;#endif
Signature · 1 line
1.7.0.6
RFC PATCH 2/23 rcu: add __rcu API for later sparse checking
In reply to Paul E. McKenneyThis commit defines an __rcu API, but provides only vacuous definitions for it. This breaks dependencies among most of the subsequent patches, allowing them to reach mainline asynchronously via whatever trees are appropriate.
--- include/linux/compiler.h | 2 ++ 1 files changed, 2 insertions(+), 0 deletions(-)
include/linux/compiler.h +2 -0
@@ -16,6 +16,7 @@# define __release (x) __context__(x,-1)# define __cond_lock (x,c) ((c) ? ({ __acquire(x); 1; }) : 0)# define __percpu __attribute__((noderef, address_space(3))) +# define __rcuextern void __chk_user_ptr (const volatile void __user *);extern void __chk_io_ptr (const volatile void __iomem *);#else
@@ -34,6 +35,7 @@ extern void __chk_io_ptr(const volatile void __iomem *);# define __release (x) (void)0# define __cond_lock (x,c) (c)# define __percpu +# define __rcu#endif #ifdef __KERNEL__
Signature · 1 line
1.7.0.6
RFC PATCH 4/23 net: Make accesses to ->br_port safe for sparse RCU
In reply to Paul E. McKenneyThe new versions of the rcu_dereference() APIs requires that any pointers passed to one of these APIs be fully defined. The ->br_port field in struct net_device points to a struct net_bridge_port, which is an incomplete type. This commit therefore changes ->br_port to be a void*, and introduces a br_port() helper function to convert the type to struct net_bridge_port, and applies this new helper function where required.
--- include/linux/if_bridge.h | 3 +++ net/bridge/br_fdb.c | 2 +- net/bridge/br_private.h | 8 ++++++++ net/bridge/netfilter/ebt_redirect.c | 2 +- net/bridge/netfilter/ebt_ulog.c | 4 ++-- net/bridge/netfilter/ebtables.c | 4 ++-- net/netfilter/nfnetlink_log.c | 4 ++-- net/netfilter/nfnetlink_queue.c | 4 ++-- 8 files changed, 21 insertions(+), 10 deletions(-)
include/linux/if_bridge.h +3 -0
@@ -101,6 +101,9 @@ struct __fdb_entry {#include <linux/netdevice.h> +/* br_handle_frame_hook() needs the following forward declaration. */ +struct net_bridge_port ; +extern void brioctl_set (int (*ioctl_hook)(struct net *,unsigned int ,void __user *));extern struct sk_buff *(*br_handle_frame_hook)(struct net_bridge_port *p,struct sk_buff *skb);
net/bridge/br_fdb.c +1 -1
@@ -246,7 +246,7 @@ int br_fdb_test_addr(struct net_device *dev, unsigned char *addr)return 0 ;rcu_read_lock (); - fdb =__br_fdb_get (dev->br_port->br, addr); + fdb =__br_fdb_get (br_port (dev)->br, addr); ret = fdb && fdb->dst->dev != dev && fdb->dst->state ==BR_STATE_FORWARDING ;rcu_read_unlock ();
net/bridge/br_private.h +8 -0
@@ -229,6 +229,14 @@ static inline int br_is_root_bridge(const struct net_bridge *br)return !memcmp (&br->bridge_id, &br->designated_root,8 ); } +static inline struct net_bridge_port *br_port (const struct net_device *dev) +{ +if (!dev) +return NULL ; + +return rcu_dereference (dev->br_port); +} +/* br_device.c */ extern void br_dev_setup (struct net_device *dev); extern netdev_tx_t br_dev_xmit(struct sk_buff *skb,
net/bridge/netfilter/ebt_redirect.c +1 -1
@@ -25,7 +25,7 @@ ebt_redirect_tg(struct sk_buff *skb, const struct xt_target_param *par)if (par->hooknum !=NF_BR_BROUTING )memcpy (eth_hdr (skb)->h_dest, - par->in->br_port->br->dev->dev_addr,ETH_ALEN ); +br_port (par->in)->br->dev->dev_addr,ETH_ALEN );else memcpy (eth_hdr (skb)->h_dest, par->in->dev_addr,ETH_ALEN ); skb->pkt_type =PACKET_HOST ;
net/bridge/netfilter/ebt_ulog.c +2 -2
@@ -183,7 +183,7 @@ static void ebt_ulog_packet(unsigned int hooknr, const struct sk_buff *skb,strcpy (pm->physindev, in->name);/* If in isn't a bridge, then physindev==indev */ if (in->br_port) -strcpy (pm->indev, in->br_port->br->dev->name); +strcpy (pm->indev,br_port (in)->br->dev->name);else strcpy (pm->indev, in->name); }else
@@ -192,7 +192,7 @@ static void ebt_ulog_packet(unsigned int hooknr, const struct sk_buff *skb,if (out) {/* If out exists, then out is a bridge port */ strcpy (pm->physoutdev, out->name); -strcpy (pm->outdev, out->br_port->br->dev->name); +strcpy (pm->outdev,br_port (out)->br->dev->name); }else pm->outdev[0 ] = pm->physoutdev[0 ] ='\0' ;
net/bridge/netfilter/ebtables.c +2 -2
@@ -141,10 +141,10 @@ ebt_basic_match(const struct ebt_entry *e, const struct ethhdr *h,if (FWINV2 (ebt_dev_check (e->out, out),EBT_IOUT ))return 1 ;if ((!in || !in->br_port) ?0 :FWINV2 (ebt_dev_check ( - e->logical_in, in->br_port->br->dev),EBT_ILOGICALIN )) + e->logical_in,br_port (in)->br->dev),EBT_ILOGICALIN ))return 1 ;if ((!out || !out->br_port) ?0 :FWINV2 (ebt_dev_check ( - e->logical_out, out->br_port->br->dev),EBT_ILOGICALOUT )) + e->logical_out,br_port (out)->br->dev),EBT_ILOGICALOUT ))return 1 ;if (e->bitmask &EBT_SOURCEMAC ) {
net/netfilter/nfnetlink_log.c +2 -2
@@ -404,7 +404,7 @@ __build_packet_message(struct nfulnl_instance *inst, htonl(indev->ifindex));/* this is the bridge group "brX" */ NLA_PUT_BE32 (inst->skb,NFULA_IFINDEX_INDEV , -htonl (indev->br_port->br->dev->ifindex)); +htonl (br_port (indev)->br->dev->ifindex)); }else {/* Case 2: indev is bridge group, we need to look for * physical device (when called from ipv4) */
@@ -431,7 +431,7 @@ __build_packet_message(struct nfulnl_instance *inst, htonl(outdev->ifindex));/* this is the bridge group "brX" */ NLA_PUT_BE32 (inst->skb,NFULA_IFINDEX_OUTDEV , -htonl (outdev->br_port->br->dev->ifindex)); +htonl (br_port (outdev)->br->dev->ifindex)); }else {/* Case 2: indev is a bridge group, we need to look * for physical device (when called from ipv4) */
net/netfilter/nfnetlink_queue.c +2 -2
@@ -298,7 +298,7 @@ nfqnl_build_packet_message(struct nfqnl_instance *queue, htonl(indev->ifindex));/* this is the bridge group "brX" */ NLA_PUT_BE32 (skb,NFQA_IFINDEX_INDEV , -htonl (indev->br_port->br->dev->ifindex)); +htonl (br_port (indev)->br->dev->ifindex)); }else {/* Case 2: indev is bridge group, we need to look for * physical device (when called from ipv4) */
@@ -323,7 +323,7 @@ nfqnl_build_packet_message(struct nfqnl_instance *queue, htonl(outdev->ifindex));/* this is the bridge group "brX" */ NLA_PUT_BE32 (skb,NFQA_IFINDEX_OUTDEV , -htonl (outdev->br_port->br->dev->ifindex)); +htonl (br_port (outdev)->br->dev->ifindex)); }else {/* Case 2: outdev is bridge group, we need to look for * physical output device (when called from ipv4) */
Signature · 1 line
1.7.0.6
RFC PATCH 15/23 radix-tree: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/radix-tree.h | 4 +++- lib/radix-tree.c | 2 +- 2 files changed, 4 insertions(+), 2 deletions(-)
include/linux/radix-tree.h +3 -1
@@ -47,6 +47,8 @@ static inline void *radix_tree_indirect_to_ptr(void *ptr) {return (void *)((unsigned long )ptr & ~RADIX_TREE_INDIRECT_PTR ); } +#define radix_tree_indirect_to_ptr (ptr) \ + radix_tree_indirect_to_ptr((void __force *)(ptr))static inline int radix_tree_is_indirect_ptr (void *ptr) {
@@ -61,7 +63,7 @@ static inline int radix_tree_is_indirect_ptr(void *ptr)struct radix_tree_root {unsigned int height;gfp_t gfp_mask; -struct radix_tree_node *rnode; +struct radix_tree_node __rcu *rnode; };#define RADIX_TREE_INIT (mask) { \
lib/radix-tree.c +1 -1
@@ -49,7 +49,7 @@ struct radix_tree_node {unsigned int height;/* Height from the bottom */ unsigned int count;struct rcu_head rcu_head; -void *slots[RADIX_TREE_MAP_SIZE ]; +void __rcu *slots[RADIX_TREE_MAP_SIZE ];unsigned long tags[RADIX_TREE_MAX_TAGS ][RADIX_TREE_TAG_LONGS ]; };
Signature · 1 line
1.7.0.6
RFC PATCH 11/23 nfs: __rcu annotations
In reply to Paul E. McKenneyFrom: Arnd Bergmann <[email protected]>
--- include/linux/nfs_fs.h | 2 +- include/linux/sunrpc/auth_gss.h | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)
include/linux/nfs_fs.h +1 -1
@@ -178,7 +178,7 @@ struct nfs_inode {struct nfs4_cached_acl *nfs4_acl;/* NFSv4 state */ struct list_head open_states; -struct nfs_delegation *delegation; +struct nfs_delegation __rcu *delegation;fmode_t delegation_state;struct rw_semaphore rwsem;#endif /* CONFIG_NFS_V4*/
include/linux/sunrpc/auth_gss.h +2 -2
@@ -69,7 +69,7 @@ struct gss_cl_ctx {enum rpc_gss_proc gc_proc;u32 gc_seq;spinlock_t gc_seq_lock; -struct gss_ctx *gc_gss_ctx; +struct gss_ctx __rcu *gc_gss_ctx;struct xdr_netobj gc_wire_ctx;u32 gc_win;unsigned long gc_expiry;
@@ -80,7 +80,7 @@ struct gss_upcall_msg;struct gss_cred {struct rpc_cred gc_base;enum rpc_gss_svc gc_service; -struct gss_cl_ctx *gc_ctx; +struct gss_cl_ctx __rcu *gc_ctx;struct gss_upcall_msg *gc_upcall;unsigned char gc_machine_cred :1 ; };
Signature · 1 line
1.7.0.6
RFC PATCH 1/23 rcu: add an rcu_dereference_index_check()
In reply to Paul E. McKenneyThe sparse RCU-pointer checking relies on type magic that dereferences the pointer in question. This does not work if the pointer is in fact an array index. This commit therefore supplies a new RCU API that omits the sparse checking to continue to support rcu_dereference() on integers.
--- include/linux/rcupdate.h | 33 +++++++++++++++++++++++++++++++++ 1 files changed, 33 insertions(+), 0 deletions(-)
include/linux/rcupdate.h +33 -0
@@ -560,4 +560,37 @@ static inline void debug_rcu_head_unqueue(struct rcu_head *head) }#endif /* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ +#ifndef CONFIG_PROVE_RCU +#define __do_rcu_dereference_check (c) do { } while (0) +#endif /* #ifdef CONFIG_PROVE_RCU */ + +#define __rcu_dereference_index_check (p, c) \ + ({ \ + typeof(p) _________p1 = ACCESS_ONCE(p); \ + __do_rcu_dereference_check(c); \ + smp_read_barrier_depends(); \ + (_________p1); \ + }) + +/** +* rcu_dereference_index_check() - rcu_dereference for indices with debug checking +* @p: The pointer to read, prior to dereferencing +* @c: The conditions under which the dereference will take place +* +* Similar to rcu_dereference_check(), but omits the sparse checking. +* This allows rcu_dereference_index_check() to be used on integers, +* which can then be used as array indices. Attempting to use +* rcu_dereference_check() on an integer will give compiler warnings +* because the sparse address-space mechanism relies on dereferencing +* the RCU-protected pointer. Dereferencing integers is not something +* that even gcc will put up with. +* +* Note that this function does not implicitly check for RCU read-side +* critical sections. If this function gains lots of uses, it might +* make sense to provide versions for each flavor of RCU, but it does +* not make sense as of early 2010. +*/ +#define rcu_dereference_index_check (p, c) \ + __rcu_dereference_index_check((p), (c)) +#endif /* __LINUX_RCUPDATE_H */
Signature · 1 line
1.7.0.6
RFC PATCH 6/23 rcu: define __rcu address space modifier for sparse
In reply to Paul E. McKenneyThis commit provides definitions for the __rcu annotation defined earlier. This annotation permits sparse to check for correct use of RCU-protected pointers. If a pointer that is annotated with __rcu is accessed directly (as opposed to via rcu_dereference(), rcu_assign_pointer(), or one of their variants), sparse can be made to complain. To enable such complaints, use the new default-disabled CONFIG_SPARSE_RCU_POINTER kernel configuration option. Please note that these sparse complaints are intended to be a debugging aid, -not- a code-style-enforcement mechanism.
There are special rcu_dereference_protected() and rcu_access_pointer() accessors for use when RCU read-side protection is not required, for example, when no other CPU has access to the data structure in question or while the current CPU hold the update-side lock.
This patch also updates a number of docbook comments that were showing their age.
--- include/linux/compiler.h | 4 + include/linux/rcupdate.h | 352 ++++++++++++++++++++++++++++------------------ include/linux/srcu.h | 27 +++- kernel/rcupdate.c | 6 +- lib/Kconfig.debug | 13 ++ 5 files changed, 257 insertions(+), 145 deletions(-)
include/linux/compiler.h +4 -0
@@ -16,7 +16,11 @@# define __release (x) __context__(x,-1)# define __cond_lock (x,c) ((c) ? ({ __acquire(x); 1; }) : 0)# define __percpu __attribute__((noderef, address_space(3))) +#ifdef CONFIG_SPARSE_RCU_POINTER +# define __rcu __attribute__((noderef, address_space(4))) +#else # define __rcu +#endif extern void __chk_user_ptr (const volatile void __user *);extern void __chk_io_ptr (const volatile void __iomem *);#else
include/linux/rcupdate.h +213 -139
477 lines not on this page · raw message
include/linux/srcu.h +23 -4
@@ -108,12 +108,31 @@ static inline int srcu_read_lock_held(struct srcu_struct *sp)#endif /* #else #ifdef CONFIG_DEBUG_LOCK_ALLOC */ /** -* srcu_dereference - fetch SRCU-protected pointer with checking +* srcu_dereference_check - fetch SRCU-protected pointer for later dereferencing +* @p: the pointer to fetch and protect for later dereferencing +* @sp: pointer to the srcu_struct, which is used to check that we +* really are in an SRCU read-side critical section. +* @c: condition to check for update-side use * -* Makes rcu_dereference_check() do the dirty work. +* If PROVE_RCU is enabled, invoking this outside of an RCU read-side +* critical section will result in an RCU-lockdep splat, unless @c evaluates +* to 1. The @c argument will normally be a logical expression containing +* lockdep_is_held() calls. */ -#define srcu_dereference (p, sp) \ - rcu_dereference_check(p, srcu_read_lock_held(sp)) +#define srcu_dereference_check (p, sp, c) \ + __rcu_dereference_check((p), srcu_read_lock_held(sp) || (c), __rcu) + +/** +* srcu_dereference - fetch SRCU-protected pointer for later dereferencing +* @p: the pointer to fetch and protect for later dereferencing +* @sp: pointer to the srcu_struct, which is used to check that we +* really are in an SRCU read-side critical section. +* +* Makes rcu_dereference_check() do the dirty work. If PROVE_RCU +* is enabled, invoking this outside of an RCU read-side critical +* section will result in an RCU-lockdep splat. +*/ +#define srcu_dereference (p, sp) srcu_dereference_check((p), (sp), 0)/** * srcu_read_lock - register a new reader for an SRCU-protected structure.
kernel/rcupdate.c +4 -2
@@ -73,12 +73,14 @@ int debug_lockdep_rcu_enabled(void)EXPORT_SYMBOL_GPL (debug_lockdep_rcu_enabled);/** -* rcu_read_lock_bh_held - might we be in RCU-bh read-side critical section? +* rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section? * * Check for bottom half being disabled, which covers both the * CONFIG_PROVE_RCU and not cases. Note that if someone uses * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled) -* will show the situation. +* will show the situation. This is useful for debug checks in functions +* that require that they be called within an RCU read-side critical +* section. * * Check debug_lockdep_rcu_enabled() to prevent false positives during boot. */
lib/Kconfig.debug +13 -0
@@ -528,6 +528,19 @@ config PROVE_RCU_REPEATEDLY disabling, allowing multiple RCU-lockdep warnings to be printed on a single reboot. +config SPARSE_RCU_POINTER + bool "RCU debugging: sparse-based checks for pointer usage" + default n + help + This feature enables the __rcu sparse annotation for + RCU-protected pointers. This annotation will cause sparse + to flag any non-RCU used of annotated pointers. This can be + helpful when debugging RCU usage. Please note that this feature + is not intended to enforce code cleanliness; it is instead merely + a debugging aid. + + Say Y to make sparse flag questionable use of RCU-protected pointers + Say N if you are unsure. config LOCKDEP
Signature · 1 line
1.7.0.6
Re: [PATCH RFC tip/core/rcu 04/23] net: Make accesses to ->br_port safe for sparse RCU
In reply to Paul E. McKenneyQuoted 18 lines · On Wed, 12 May 2010 14:33:23 -0700 "Paul E. McKenney"…
net/bridge/br_fdb.c +1
-1@@ -246,7 +246,7 @@ int br_fdb_test_addr(struct net_device *dev, unsigned char *addr)return 0 ;rcu_read_lock (); - fdb =__br_fdb_get (dev->br_port->br, addr); + fdb =__br_fdb_get (br_port (dev)->br, addr); ret = fdb && fdb->dst->dev != dev && fdb->dst->state ==BR_STATE_FORWARDING ;rcu_read_unlock ();net/bridge/br_private.h +0
-0
net/bridge/br_private.h static inline int br_is_root_bridge(const struct net_bridge *br) · line 238
Quoted 10 lines
net/bridge/br_private.h +7
-0@@ -229,6 +229,14 @@ static inline int br_is_root_bridge(const struct net_bridge *br)return !memcmp (&br->bridge_id, &br->designated_root,8 ); } +static inline struct net_bridge_port *br_port (const struct net_device *dev) +{ +if (!dev) +return NULL ; + +return rcu_dereference (dev->br_port); +}
Looks like this is wrapping existing problems, and hurting not helping.
Why introduce a wrapper that could return NULL and not check the result?
I would rather that:
1. dev should never be null in this cases so the first if() is unnecessary, and confuses the semantics.
2. don't use wrapper br_port()
3. have callers check that rcu_dereference(dev->br_port) did not return NULL.
If they derefernce does return NULL, then it means other CPU has started tear down and this CPU should just go home quietly.
Michael S. Tsirkin
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 7 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Also add
rcu_dreference_protected()for code paths where locks are held.Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc "Michael S. Tsirkin" <[email protected]>
Maybe long lines can be fixed? Otherwise looks ok.
Quoted 118 lines
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)drivers/vhost/net.c +8
-3@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;drivers/vhost/vhost.c +8
-6@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);drivers/vhost/vhost.h +2
-2@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;Signature · 1 line
1.7.0.6
Re: [PATCH RFC tip/core/rcu 04/23] net: Make accesses to ->br_port safe for sparse RCU
In reply to Stephen HemmingerQuoted 34 lines · Stephen Hemminger
Quoted 18 lines · On Wed, 12 May 2010 14:33:23 -0700 "Paul E. McKenney"…
net/bridge/br_fdb.c +1
-1@@ -246,7 +246,7 @@ int br_fdb_test_addr(struct net_device *dev, unsigned char *addr)return 0 ;rcu_read_lock (); - fdb =__br_fdb_get (dev->br_port->br, addr); + fdb =__br_fdb_get (br_port (dev)->br, addr); ret = fdb && fdb->dst->dev != dev && fdb->dst->state ==BR_STATE_FORWARDING ;rcu_read_unlock ();net/bridge/br_private.h +0
-0net/bridge/br_private.h static inline int br_is_root_bridge(const struct net_bridge *br) · line 238
Quoted 10 lines
net/bridge/br_private.h +7
-0@@ -229,6 +229,14 @@ static inline int br_is_root_bridge(const struct net_bridge *br)return !memcmp (&br->bridge_id, &br->designated_root,8 ); } +static inline struct net_bridge_port *br_port (const struct net_device *dev) +{ +if (!dev) +return NULL ; + +return rcu_dereference (dev->br_port); +}Looks like this is wrapping existing problems, and hurting not helping.
Quoted 2 lines
Why introduce a wrapper that could return NULL and not check the result?
Fair point!
Quoted 8 lines
I would rather that:
1. dev should never be null in this cases so the firstif()is unnecessary, and confuses the semantics.
2. don't use wrapperbr_port()
3. have callers check thatrcu_dereference(dev->br_port)did not return NULL.
If they derefernce does return NULL, then it means other CPU has started tear down and this CPU should just go home quietly.
OK.
The reason for br_port() is to allow ->br_port to be a void*. If we eliminate br_port(), then it is necessary to make the definition of the struct net_bridge_port available everywhere that ->br_port is given to rcu_dereference(). The reason for this is that Arnd's sparse-based RCU checking code uses __rcu to tag the data pointed to by an RCU-protected pointer. This in turn means that rcu_dereference() and friends must now have access to the pointed-to type, as is done in patch 6 in this series.
One way to make struct net_bridge_port available is to put:
#include "../../net/bridge/br_private.h"
in include/linux/netdevice.h.
However, when I try this, I get lots of build errors, which was what led us to the path of making ->br_port be a void*, thus requiring the br_port() helper function in cases where the caller needs the underlying type.
What should we be doing instead?
Thanx, Paul
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Michael S. TsirkinQuoted 10 lines · Michael S. Tsirkin
Quoted 7 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Also add
rcu_dreference_protected()for code paths where locks are held.Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc "Michael S. Tsirkin" <[email protected]>Maybe long lines can be fixed? Otherwise looks ok.
Done. I introduced locals to make it fit.
One other thing... We need some API that says "we are running in the context of a work queue." Otherwise, we will get false positives when called in a work queue without locks held.
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument rcu_read_lock() and friends.
Thanx, Paul
Quoted 118 lines
Quoted 118 lines
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)drivers/vhost/net.c +8
-3@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;drivers/vhost/vhost.c +8
-6@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);drivers/vhost/vhost.h +2
-2@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;Signature · 1 line
1.7.0.6
Re: [PATCH RFC tip/core/rcu 04/23] net: Make accesses to ->br_port safe for sparse RCU
In reply to Paul E. McKenneyQuoted 51 lines · On Wed, 12 May 2010 15:35:25 -0700 "Paul E. McKenney"…
Quoted 34 lines · Stephen Hemminger
Quoted 18 lines · On Wed, 12 May 2010 14:33:23 -0700 "Paul E. McKenney"…
net/bridge/br_fdb.c +1
-1@@ -246,7 +246,7 @@ int br_fdb_test_addr(struct net_device *dev, unsigned char *addr)return 0 ;rcu_read_lock (); - fdb =__br_fdb_get (dev->br_port->br, addr); + fdb =__br_fdb_get (br_port (dev)->br, addr); ret = fdb && fdb->dst->dev != dev && fdb->dst->state ==BR_STATE_FORWARDING ;rcu_read_unlock ();net/bridge/br_private.h +0
-0net/bridge/br_private.h static inline int br_is_root_bridge(const struct net_bridge *br) · line 238
Quoted 10 lines
net/bridge/br_private.h +7
-0@@ -229,6 +229,14 @@ static inline int br_is_root_bridge(const struct net_bridge *br)return !memcmp (&br->bridge_id, &br->designated_root,8 ); } +static inline struct net_bridge_port *br_port (const struct net_device *dev) +{ +if (!dev) +return NULL ; + +return rcu_dereference (dev->br_port); +}Looks like this is wrapping existing problems, and hurting not helping.
Quoted 2 lines
Why introduce a wrapper that could return NULL and not check the result?
Fair point!
Quoted 8 lines
I would rather that:
1. dev should never be null in this cases so the firstif()is unnecessary, and confuses the semantics.
2. don't use wrapperbr_port()
3. have callers check thatrcu_dereference(dev->br_port)did not return NULL.
If they derefernce does return NULL, then it means other CPU has started tear down and this CPU should just go home quietly.OK.
Quoted 8 lines
The reason for
br_port()is to allow ->br_port to be a void*. If we eliminatebr_port(), then it is necessary to make the definition of the structnet_bridge_portavailable everywhere that ->br_port is given torcu_dereference(). The reason for this is that Arnd's sparse-based RCU checking code uses__rcuto tag the data pointed to by an RCU-protected pointer. This in turn means thatrcu_dereference()and friends must now have access to the pointed-to type, as is done in patch 6 in this series.
Then ok. leave the wrapper, but get rid of the !dev part.
I can do it if you want.
Still don't like changing working code to conform to code checking tools. Especially when code checking tool is missing bad RCU usage that already exists (like this case). It is a big problem if code assumes rcu_deref always returns non NULL.
Re: [PATCH RFC tip/core/rcu 04/23] net: Make accesses to ->br_port safe for sparse RCU
Quoted 63 lines · Stephen Hemminger
Quoted 51 lines · On Wed, 12 May 2010 15:35:25 -0700 "Paul E. McKenney"…
Quoted 34 lines · Stephen Hemminger
Quoted 18 lines · On Wed, 12 May 2010 14:33:23 -0700 "Paul E. McKenney"…
net/bridge/br_fdb.c +1
-1@@ -246,7 +246,7 @@ int br_fdb_test_addr(struct net_device *dev, unsigned char *addr)return 0 ;rcu_read_lock (); - fdb =__br_fdb_get (dev->br_port->br, addr); + fdb =__br_fdb_get (br_port (dev)->br, addr); ret = fdb && fdb->dst->dev != dev && fdb->dst->state ==BR_STATE_FORWARDING ;rcu_read_unlock ();net/bridge/br_private.h +0
-0net/bridge/br_private.h static inline int br_is_root_bridge(const struct net_bridge *br) · line 238
Quoted 10 lines
net/bridge/br_private.h +7
-0@@ -229,6 +229,14 @@ static inline int br_is_root_bridge(const struct net_bridge *br)return !memcmp (&br->bridge_id, &br->designated_root,8 ); } +static inline struct net_bridge_port *br_port (const struct net_device *dev) +{ +if (!dev) +return NULL ; + +return rcu_dereference (dev->br_port); +}Looks like this is wrapping existing problems, and hurting not helping.
Quoted 2 lines
Why introduce a wrapper that could return NULL and not check the result?
Fair point!
Quoted 8 lines
I would rather that:
1. dev should never be null in this cases so the firstif()is unnecessary, and confuses the semantics.
2. don't use wrapperbr_port()
3. have callers check thatrcu_dereference(dev->br_port)did not return NULL.
If they derefernce does return NULL, then it means other CPU has started tear down and this CPU should just go home quietly.OK.
Quoted 8 lines
The reason for
br_port()is to allow ->br_port to be a void*. If we eliminatebr_port(), then it is necessary to make the definition of the structnet_bridge_portavailable everywhere that ->br_port is given torcu_dereference(). The reason for this is that Arnd's sparse-based RCU checking code uses__rcuto tag the data pointed to by an RCU-protected pointer. This in turn means thatrcu_dereference()and friends must now have access to the pointed-to type, as is done in patch 6 in this series.
Quoted 3 lines
Then ok. leave the wrapper, but get rid of the !dev part.
I can do it if you want.
Done!
I would normally accept your offer in a heartbeat, but there are dependencies among the patches. :-(
Quoted 4 lines
Still don't like changing working code to conform to code checking tools. Especially when code checking tool is missing bad RCU usage that already exists (like this case). It is a big problem if code assumes
rcu_derefalways returns non NULL.
Indeed, I would have liked some way of making this work without having to make rcu_dereference() know about the pointed-too type. I have not yet been able to find one, though. :-(
Thanx, Paul
Michael S. Tsirkin
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 18 lines · Paul E. McKenney
Quoted 10 lines · Michael S. Tsirkin
Quoted 7 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Also add
rcu_dreference_protected()for code paths where locks are held.Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc "Michael S. Tsirkin" <[email protected]>Maybe long lines can be fixed? Otherwise looks ok.
Done. I introduced locals to make it fit.
One other thing... We need some API that says "we are running in the context of a work queue." Otherwise, we will get false positives when called in a work queue without locks held.
Quoted 5 lines
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Thanx, Paul
And we would want some way to tag the flushes ... not easy. Let's start with switching to raw to avoid the errors.
Quoted 118 lines
Quoted 118 lines
Quoted 118 lines
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)drivers/vhost/net.c +8
-3@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;drivers/vhost/vhost.c +8
-6@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);drivers/vhost/vhost.h +2
-2@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;Signature · 1 line
1.7.0.6
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 19 lines · Michael S. Tsirkin
Quoted 18 lines · Paul E. McKenney
Quoted 10 lines · Michael S. Tsirkin
Quoted 7 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Also add
rcu_dreference_protected()for code paths where locks are held.Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc "Michael S. Tsirkin" <[email protected]>Maybe long lines can be fixed? Otherwise looks ok.
Done. I introduced locals to make it fit.
One other thing... We need some API that says "we are running in the context of a work queue." Otherwise, we will get false positives when called in a work queue without locks held.
Quoted 8 lines
Quoted 5 lines
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Thanx, Paul
And we would want some way to tag the flushes ... not easy. Let's start with switching to raw to avoid the errors.
The flushes turn out to be a non-problem for the moment due to the fact that there is no syntactic way to tie them to the read-side critical sections.
On the read-side critical sections, I would not be above looking at the ->comm field and checking for the possible command names of the workqueue kernel threads. Might get some false negatives due to choices of user command names, but that is OK. We just need a high probability of catching errors, not a certainty. ;-)
Thanx, Paul
Quoted 118 lines
Quoted 118 lines
Quoted 118 lines
Quoted 118 lines
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)drivers/vhost/net.c +8
-3@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;drivers/vhost/vhost.c +8
-6@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);drivers/vhost/vhost.h +2
-2@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;Signature · 1 line
1.7.0.6
Michael S. Tsirkin
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 6 lines · Paul E. McKenney
Quoted 5 lines · Michael S. Tsirkin
Quoted 4 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Also add
rcu_dreference_protected()for code paths where locks are held.
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc "Michael S. Tsirkin" <[email protected]>Maybe long lines can be fixed? Otherwise looks ok.
Done. I introduced locals to make it fit.
Just pls check this does not lead to unused variable warnings when the whole lockdep is off.
Quoted 128 lines
One other thing... We need some API that says "we are running in the context of a work queue." Otherwise, we will get false positives when called in a work queue without locks held.
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Thanx, Paul
Quoted 118 lines
Quoted 118 lines
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)drivers/vhost/net.c +8
-3@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;drivers/vhost/vhost.c +8
-6@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);drivers/vhost/vhost.h +2
-2@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;Signature · 1 line
1.7.0.6
Re: [PATCH RFC tip/core/rcu 17/23] input: __rcu annotations
In reply to Paul E. McKenneyQuoted 4 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>
Please merge with the rest of RCU annotations.
Signature · 1 line
Dmitry
Re: [PATCH RFC tip/core/rcu 09/23] credentials: rcu annotation
In reply to Paul E. McKenneyQuoted 8 lines · Paul E. McKenney
Subject: [PATCH RFC tip/core/rcu 09/23] credentials: rcu annotation
From: Arnd Bergmann <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc Peter Zijlstra <[email protected]>Cc Ingo Molnar <[email protected]>
Re: [PATCH RFC tip/core/rcu 10/23] keys: __rcu annotations
In reply to Paul E. McKenneyQuoted 6 lines · Paul E. McKenney
Subject: [PATCH RFC tip/core/rcu 10/23] keys:
__rcuannotationsFrom: Arnd Bergmann <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>
Peter Zijlstra
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.
Re: [PATCH RFC tip/core/rcu 18/23] net/netfilter: __rcu annotations
In reply to Paul E. McKenneyQuoted 7 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc Patrick McHardy <[email protected]>Cc "David S. Miller" <[email protected]>Cc Eric Dumazet <[email protected]>
Quoted 10 lines
--- include/net/netfilter/nf_conntrack.h | 2 +- net/ipv4/netfilter/nf_nat_core.c | 2 +- net/netfilter/core.c | 2 +- net/netfilter/nf_conntrack_ecache.c | 4 ++-- net/netfilter/nf_conntrack_extend.c | 2 +- net/netfilter/nf_conntrack_proto.c | 4 ++-- net/netfilter/nf_log.c | 2 +- net/netfilter/nf_queue.c | 2 +- 8 files changed, 10 insertions(+), 10 deletions(-)
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Peter ZijlstraQuoted 8 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.
Thank you, Peter!!!
Thanx, Paul
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Michael S. TsirkinQuoted 11 lines · Michael S. Tsirkin
Quoted 10 lines · Paul E. McKenney
Quoted 9 lines · Michael S. Tsirkin
Quoted 7 lines · Paul E. McKenney
From: Arnd Bergmann <[email protected]>
Also add
rcu_dreference_protected()for code paths where locks are held.Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc "Michael S. Tsirkin" <[email protected]>
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
Maybe long lines can be fixed? Otherwise looks ok.
Done. I introduced locals to make it fit.
Just pls check this does not lead to unused variable warnings when the whole lockdep is off.
Will do! It should not, because the assignment happens independently of lockdep, but I will nevertheless build both ways.
I am also merging in Peter Zijlstra's suggestion of leveraging the existing lockdep on workqueues.
Thanx, Paul
Quoted 128 lines
Quoted 128 lines
One other thing... We need some API that says "we are running in the context of a work queue." Otherwise, we will get false positives when called in a work queue without locks held.
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Thanx, Paul
Quoted 118 lines
Quoted 118 lines
--- drivers/vhost/net.c | 11 ++++++++--- drivers/vhost/vhost.c | 14 ++++++++------ drivers/vhost/vhost.h | 4 ++-- 3 files changed, 18 insertions(+), 11 deletions(-)drivers/vhost/net.c +8
-3@@ -364,7 +364,10 @@ static void vhost_net_disable_vq(struct vhost_net *n,static void vhost_net_enable_vq (struct vhost_net *n,struct vhost_virtqueue *vq) { -struct socket *sock = vq->private_data; +struct socket *sock; + + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (!sock)return ;if (vq == n->vqs +VHOST_NET_VQ_TX ) {@@ -380,7 +383,8 @@ static struct socket *vhost_net_stop_vq(struct vhost_net *n,struct socket *sock;mutex_lock (&vq->mutex); - sock = vq->private_data; + sock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));vhost_net_disable_vq (n, vq);rcu_assign_pointer (vq->private_data,NULL );mutex_unlock (&vq->mutex);@@ -518,7 +522,8 @@ static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd) }/* start polling new socket */ - oldsock = vq->private_data; + oldsock =rcu_dereference_protected (vq->private_data, +lockdep_is_held (&vq->mutex));if (sock == oldsock) goto done;drivers/vhost/vhost.c +8
-6@@ -180,7 +180,7 @@ long vhost_dev_reset_owner(struct vhost_dev *dev)vhost_dev_cleanup (dev); memory->nregions =0 ; - dev->memory = memory; +RCU_INIT_POINTER (dev->memory, memory);return 0 ; }@@ -212,8 +212,9 @@ void vhost_dev_cleanup(struct vhost_dev *dev)fput (dev->log_file); dev->log_file =NULL ;/* No one will access memory at this point */ -kfree (dev->memory); - dev->memory =NULL ; +kfree (rcu_dereference_protected (dev->memory, +lockdep_is_held (&dev->mutex))); +RCU_INIT_POINTER (dev->memory,NULL );if (dev->mm)mmput (dev->mm); dev->mm =NULL ;@@ -294,14 +295,14 @@ static int vq_access_ok(unsigned int num, /* Caller should have device mutex but not vq mutex */ int vhost_log_access_ok(struct vhost_dev *dev) { - return memory_access_ok(dev, dev->memory, 1); + return memory_access_ok(dev, rcu_dereference_protected(dev->memory, lockdep_is_held(&dev->mutex)), 1); } /* Verify access for write logging. */ /* Caller should have vq mutex and device mutex */ static int vq_log_access_ok(struct vhost_virtqueue *vq, void __user *log_base) { - return vq_memory_access_ok(log_base, vq->dev->memory, + return vq_memory_access_ok(log_base, rcu_dereference_protected(vq->dev->memory, lockdep_is_held(&dev->mutex)), vhost_has_feature(vq->dev, VHOST_F_LOG_ALL)) && (!vq->log_used || log_access_ok(log_base, vq->log_addr, sizeof *vq->used +@@ -342,7 +343,8 @@ static long vhost_set_memory(struct vhost_dev *d, struct vhost_memory __user *m)if (!memory_access_ok (d, newmem,vhost_has_feature (d,VHOST_F_LOG_ALL )))return -EFAULT ; - oldmem = d->memory; + oldmem =rcu_dereference_protected (d->memory, +lockdep_is_held (&d->mutex));rcu_assign_pointer (d->memory, newmem);synchronize_rcu ();kfree (oldmem);drivers/vhost/vhost.h +2
-2@@ -92,7 +92,7 @@ struct vhost_virtqueue {* work item execution acts instead of rcu_read_lock() and the end of * work item execution acts instead of rcu_read_lock(). * Writers use virtqueue mutex. */ -void *private_data; +void __rcu *private_data;/* Log write descriptors */ void __user *log_base;struct vhost_log log[VHOST_NET_MAX_SG ];@@ -102,7 +102,7 @@ struct vhost_dev {/* Readers use RCU to access memory table pointer * log base pointer and features. * Writers use mutex below.*/ -struct vhost_memory *memory; +struct vhost_memory __rcu *memory;struct mm_struct *mm;struct mutex mutex;unsigned acked_features;Signature · 1 line
1.7.0.6
Re: [PATCH RFC tip/core/rcu 02/23] rcu: add __rcu API for later sparse checking
In reply to Paul E. McKenneyQuoted 4 lines · Paul E. McKenney
This commit defines an
__rcuAPI, but provides only vacuous definitions for it. This breaks dependencies among most of the subsequent patches, allowing them to reach mainline asynchronously via whatever trees are appropriate.
Seems like a good plan to me.
I know it's not the right time to push it but I am curious to see what, approximately, you expect a non-vacuous __rcu definition to look like. (i.e. when it's being run through sparse)
Cheers,
-Matt Helsley
Quoted 37 lines
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc Christopher Li <[email protected]>Cc Josh Triplett <[email protected]>--- include/linux/compiler.h | 2 ++ 1 files changed, 2 insertions(+), 0 deletions(-)include/linux/compiler.h +2
-0@@ -16,6 +16,7 @@# define __release (x) __context__(x,-1)# define __cond_lock (x,c) ((c) ? ({ __acquire(x); 1; }) : 0)# define __percpu __attribute__((noderef, address_space(3))) +# define __rcuextern void __chk_user_ptr (const volatile void __user *);extern void __chk_io_ptr (const volatile void __iomem *);#else @@ -34,6 +35,7 @@ extern void __chk_io_ptr(const volatile void __iomem *);# define __release (x) (void)0# define __cond_lock (x,c) (c)# define __percpu +# define __rcu#endif #ifdef __KERNEL__Signature · 7 lines
1.7.0.6 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC tip/core/rcu 02/23] rcu: add __rcu API for later sparse checking
Quoted 11 lines · Matt Helsley
Quoted 4 lines · Paul E. McKenney
This commit defines an
__rcuAPI, but provides only vacuous definitions for it. This breaks dependencies among most of the subsequent patches, allowing them to reach mainline asynchronously via whatever trees are appropriate.Seems like a good plan to me.
I know it's not the right time to push it but I am curious to see what, approximately, you expect a non-vacuous
__rcudefinition to look like. (i.e. when it's being run through sparse)
Patch 6 of this series defines the sparse magic and the rcu_dereference() checks that do this. But yes, there are dependencies among these patches.
Thanx, Paul
Quoted 40 lines
Cheers,
-Matt HelsleyQuoted 37 lines
Signed-off-by Arnd Bergmann <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Cc Christopher Li <[email protected]>Cc Josh Triplett <[email protected]>--- include/linux/compiler.h | 2 ++ 1 files changed, 2 insertions(+), 0 deletions(-)include/linux/compiler.h +2
-0@@ -16,6 +16,7 @@# define __release (x) __context__(x,-1)# define __cond_lock (x,c) ((c) ? ({ __acquire(x); 1; }) : 0)# define __percpu __attribute__((noderef, address_space(3))) +# define __rcuextern void __chk_user_ptr (const volatile void __user *);extern void __chk_io_ptr (const volatile void __iomem *);#else @@ -34,6 +35,7 @@ extern void __chk_io_ptr(const volatile void __iomem *);# define __release (x) (void)0# define __cond_lock (x,c) (c)# define __percpu +# define __rcu#endif #ifdef __KERNEL__Signature · 7 lines
1.7.0.6 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Michael S. Tsirkin
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.
Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
vhost in fact does flush_work rather than flush_workqueue, so while for now everything runs from vhost_workqueue in theory nothing would break if we use some other workqueue or even a combination thereof.
I guess when/if this happens, we could start by converting to _raw and then devise a solution.
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Signature · 1 line
MST
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.
If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU
int in_vhost_workqueue(void)
{
return in_workqueue_context(vhost_workqueue) ||
in_workqueue_context(vhost_other_workqueue) ||
in_workqueue_context(yet_another_vhost_workqueue);
}
#endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as synchronize_rcu() has no way to know what memory it is flushing, nor does flush_work().
Thanx, Paul
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 44 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU int in_vhost_workqueue(void) { return in_workqueue_context(vhost_workqueue) || in_workqueue_context(vhost_other_workqueue) || in_workqueue_context(yet_another_vhost_workqueue); } #endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Quoted 4 lines
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as
synchronize_rcu()has no way to know what memory it is flushing, nor doesflush_work().
We can think of my recent debugobjects addition as a small state machine that is described by the code that owns the objects. At each state transition, the code passes the expected state as well as the next state.
The current implementation can only keep track of a single "state" per object at once. This should be extended to be able to count the number RCU read side C.S. in flight that are accessing to an object.
We could use a hook in rcu_dereference (which knows about the object) and a hook in rcu_read_unlock (which determines the end of valid object use).
We should hook into rcu_assign_pointer() to detect RCU structure privatization. It should put these objects in a "privatized" hash table.
We should also hook into synchronize_rcu/sched() to remove the privatized structures from the privatized hash.
A hook in "kfree" (maybe a new rcu_free(void (fctptr*)(void *)) wrapper ?) would call a debugobject hook that would lookup the "privatized" hash. If it contains the object to free, we check if there are RCU read-side C.S. in flight using this object at the same time, and show an error if both are true.
Thoughts ?
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 50 lines · Mathieu Desnoyers
Quoted 44 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU int in_vhost_workqueue(void) { return in_workqueue_context(vhost_workqueue) || in_workqueue_context(vhost_other_workqueue) || in_workqueue_context(yet_another_vhost_workqueue); } #endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Quoted 4 lines
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as
synchronize_rcu()has no way to know what memory it is flushing, nor doesflush_work().
Quoted 8 lines
We can think of my recent debugobjects addition as a small state machine that is described by the code that owns the objects. At each state transition, the code passes the expected state as well as the next state.
The current implementation can only keep track of a single "state" per object at once. This should be extended to be able to count the number RCU read side
C.S. in flight that are accessing to an object.
Not a problem, as vhost doesn't use call_rcu(). So there won't be a conflict between different debugobjects views of the same memory.
Quoted 7 lines
We could use a hook in
rcu_dereference(which knows about the object) and a hook inrcu_read_unlock(which determines the end of valid object use).We should hook into
rcu_assign_pointer()to detect RCU structure privatization. It should put these objects in a "privatized" hash table.
Quoted 8 lines
We should also hook into synchronize_rcu/sched() to remove the privatized structures from the privatized hash.
A hook in "kfree" (maybe a new
rcu_free(void (fctptr*)(void *))wrapper ?) would call a debugobject hook that would lookup the "privatized" hash. If it contains the object to free, we check if there are RCU read-sideC.S. in flight using this object at the same time, and show an error if both are true.
I believe that we can't bury this into the RCU primitives, because rcu_read_unlock() doesn't know what objects were referenced in the RCU read-side critical section.
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
Thanx, Paul
Quoted 13 lines
Thoughts ?
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Michael S. Tsirkin
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 83 lines · Paul E. McKenney
Quoted 50 lines · Mathieu Desnoyers
Quoted 44 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU int in_vhost_workqueue(void) { return in_workqueue_context(vhost_workqueue) || in_workqueue_context(vhost_other_workqueue) || in_workqueue_context(yet_another_vhost_workqueue); } #endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Quoted 4 lines
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as
synchronize_rcu()has no way to know what memory it is flushing, nor doesflush_work().Quoted 8 lines
We can think of my recent debugobjects addition as a small state machine that is described by the code that owns the objects. At each state transition, the code passes the expected state as well as the next state.
The current implementation can only keep track of a single "state" per object at once. This should be extended to be able to count the number RCU read side
C.S. in flight that are accessing to an object.Not a problem, as vhost doesn't use
call_rcu(). So there won't be a conflict between different debugobjects views of the same memory.Quoted 7 lines
We could use a hook in
rcu_dereference(which knows about the object) and a hook inrcu_read_unlock(which determines the end of valid object use).We should hook into
rcu_assign_pointer()to detect RCU structure privatization. It should put these objects in a "privatized" hash table.Quoted 8 lines
We should also hook into synchronize_rcu/sched() to remove the privatized structures from the privatized hash.
A hook in "kfree" (maybe a new
rcu_free(void (fctptr*)(void *))wrapper ?) would call a debugobject hook that would lookup the "privatized" hash. If it contains the object to free, we check if there are RCU read-sideC.S. in flight using this object at the same time, and show an error if both are true.I believe that we can't bury this into the RCU primitives, because
rcu_read_unlock()doesn't know what objects were referenced in the RCU read-side critical section.
Quoted 5 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
Thanx, Paul
Well, it does not have to be freed, memory could be reused for something else. I was just saying that it would be nice to catch the class of errors which includes a missing sync.
Quoted 13 lines
Quoted 13 lines
Thoughts ?
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 56 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 55 lines · Mathieu Desnoyers
Quoted 44 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU int in_vhost_workqueue(void) { return in_workqueue_context(vhost_workqueue) || in_workqueue_context(vhost_other_workqueue) || in_workqueue_context(yet_another_vhost_workqueue); } #endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Quoted 4 lines
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as
synchronize_rcu()has no way to know what memory it is flushing, nor doesflush_work().We can think of my recent debugobjects addition as a small state machine that is described by the code that owns the objects. At each state transition, the code passes the expected state as well as the next state.
Quoted 6 lines
Quoted 3 lines
The current implementation can only keep track of a single "state" per object at once. This should be extended to be able to count the number RCU read side
C.S. in flight that are accessing to an object.Not a problem, as vhost doesn't use
call_rcu(). So there won't be a conflict between different debugobjects views of the same memory.
Not quite sure I follow you here.
Quoted 17 lines
Quoted 8 lines
We could use a hook in
rcu_dereference(which knows about the object) and a hook inrcu_read_unlock(which determines the end of valid object use).We should hook into
rcu_assign_pointer()to detect RCU structure privatization. It should put these objects in a "privatized" hash table.Quoted 8 lines
We should also hook into synchronize_rcu/sched() to remove the privatized structures from the privatized hash.
A hook in "kfree" (maybe a new
rcu_free(void (fctptr*)(void *))wrapper ?) would call a debugobject hook that would lookup the "privatized" hash. If it contains the object to free, we check if there are RCU read-sideC.S. in flight using this object at the same time, and show an error if both are true.
Quoted 3 lines
I believe that we can't bury this into the RCU primitives, because
rcu_read_unlock()doesn't know what objects were referenced in the RCU read-side critical section.
Well, if we can find a way to match a sequence of rcu_dereference performed from a thread with the following rcu_read_unlock(), then we might have the information we need. But we would have to somehow tie the debugobject context to the thread context. That sounds too complex for what we are trying to achieve here.
Quoted 4 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certain rcu_dereference always access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.
Thanks,
Mathieu
Quoted 16 lines
Thanx, Paul
Quoted 13 lines
Thoughts ?
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 57 lines · Mathieu Desnoyers
Quoted 56 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 55 lines · Mathieu Desnoyers
Quoted 44 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU int in_vhost_workqueue(void) { return in_workqueue_context(vhost_workqueue) || in_workqueue_context(vhost_other_workqueue) || in_workqueue_context(yet_another_vhost_workqueue); } #endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Quoted 4 lines
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as
synchronize_rcu()has no way to know what memory it is flushing, nor doesflush_work().We can think of my recent debugobjects addition as a small state machine that is described by the code that owns the objects. At each state transition, the code passes the expected state as well as the next state.
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
The current implementation can only keep track of a single "state" per object at once. This should be extended to be able to count the number RCU read side
C.S. in flight that are accessing to an object.Not a problem, as vhost doesn't use
call_rcu(). So there won't be a conflict between different debugobjects views of the same memory.Not quite sure I follow you here.
vhost uses only synchronize_rcu() and flush_work(). The existing debugobjects tagging would therefore be unaware of the actual object, instead tagging the rcu_head that synchronize_rcu() allocated on the stack, and being out of the picture completely in the case of flush_work().
Either way, RCU is completely unaware of exactly which structure is being pushed through a grace period, so RCU's debugobjects tagging cannot possibly conflict with any tagging that vhost does.
Quoted 20 lines
Quoted 16 lines
Quoted 7 lines
We could use a hook in
rcu_dereference(which knows about the object) and a hook inrcu_read_unlock(which determines the end of valid object use).We should hook into
rcu_assign_pointer()to detect RCU structure privatization. It should put these objects in a "privatized" hash table.Quoted 8 lines
We should also hook into synchronize_rcu/sched() to remove the privatized structures from the privatized hash.
A hook in "kfree" (maybe a new
rcu_free(void (fctptr*)(void *))wrapper ?) would call a debugobject hook that would lookup the "privatized" hash. If it contains the object to free, we check if there are RCU read-sideC.S. in flight using this object at the same time, and show an error if both are true.Quoted 3 lines
I believe that we can't bury this into the RCU primitives, because
rcu_read_unlock()doesn't know what objects were referenced in the RCU read-side critical section.
Quoted 5 lines
Well, if we can find a way to match a sequence of
rcu_dereferenceperformed from a thread with the followingrcu_read_unlock(), then we might have the information we need. But we would have to somehow tie the debugobject context to the thread context. That sounds too complex for what we are trying to achieve here.
Indeed! Especially given the fact that RCU read-side critical sections can be nested. Which rcu_dereference() calls go with which RCU read-side critical section?
Quoted 10 lines
Quoted 3 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any rcu_dereference() calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.
Thanx, Paul
Quoted 27 lines
Thanks,
Mathieu
Quoted 16 lines
Thanx, Paul
Quoted 13 lines
Thoughts ?
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.comSignature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 73 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 57 lines · Mathieu Desnoyers
Quoted 56 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 55 lines · Mathieu Desnoyers
Quoted 44 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 15 lines · Michael S. Tsirkin
Quoted 6 lines · Paul E. McKenney
Quoted 5 lines · Peter Zijlstra
Quoted 3 lines · Paul E. McKenney
Any thoughts? One approach would be to create a separate lockdep class for vhost workqueue state, similar to the approach used in instrument
rcu_read_lock()and friends.Quoted 7 lines
Quoted 3 lines
workqueue_struct::lockdep_map, its held while executing worklets.
lock_is_held(&vhost_workqueue->lockdep_map), should do as you want.Thank you, Peter!!!
Thanx, Paul
Quoted 8 lines
vhost in fact does
flush_workrather thanflush_workqueue, so while for now everything runs fromvhost_workqueuein theory nothing would break if we use some other workqueue or even a combination thereof.I guess when/if this happens, we could start by converting to
_rawand then devise a solution.If there are a small finite number of work queues involved, we can easily do something like:
#ifdef CONFIG_PROVE_RCU int in_vhost_workqueue(void) { return in_workqueue_context(vhost_workqueue) || in_workqueue_context(vhost_other_workqueue) || in_workqueue_context(yet_another_vhost_workqueue); } #endif /* #ifdef CONFIG_PROVE_RCU */Seem reasonable?
Quoted 4 lines
By the way what would be really nice is if we had a way to trap when rcu protected pointer is freed without a flush while some reader is running. Current annotation does not allow this, does it?
Quoted 4 lines
Right now, it does not, but I wonder if something like Thomas's and Mathieu's debugobjects work could be brought to bear on this problem? This would need to be implemented in vhost, as
synchronize_rcu()has no way to know what memory it is flushing, nor doesflush_work().We can think of my recent debugobjects addition as a small state machine that is described by the code that owns the objects. At each state transition, the code passes the expected state as well as the next state.
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
The current implementation can only keep track of a single "state" per object at once. This should be extended to be able to count the number RCU read side
C.S. in flight that are accessing to an object.Not a problem, as vhost doesn't use
call_rcu(). So there won't be a conflict between different debugobjects views of the same memory.Not quite sure I follow you here.
vhost uses only
synchronize_rcu()andflush_work(). The existing debugobjects tagging would therefore be unaware of the actual object, instead tagging thercu_headthatsynchronize_rcu()allocated on the stack, and being out of the picture completely in the case offlush_work().
Quoted 3 lines
Either way, RCU is completely unaware of exactly which structure is being pushed through a grace period, so RCU's debugobjects tagging cannot possibly conflict with any tagging that vhost does.
I see.
Quoted 27 lines
Quoted 21 lines
Quoted 17 lines
Quoted 8 lines
We could use a hook in
rcu_dereference(which knows about the object) and a hook inrcu_read_unlock(which determines the end of valid object use).We should hook into
rcu_assign_pointer()to detect RCU structure privatization. It should put these objects in a "privatized" hash table.Quoted 8 lines
We should also hook into synchronize_rcu/sched() to remove the privatized structures from the privatized hash.
A hook in "kfree" (maybe a new
rcu_free(void (fctptr*)(void *))wrapper ?) would call a debugobject hook that would lookup the "privatized" hash. If it contains the object to free, we check if there are RCU read-sideC.S. in flight using this object at the same time, and show an error if both are true.Quoted 3 lines
I believe that we can't bury this into the RCU primitives, because
rcu_read_unlock()doesn't know what objects were referenced in the RCU read-side critical section.Quoted 5 lines
Well, if we can find a way to match a sequence of
rcu_dereferenceperformed from a thread with the followingrcu_read_unlock(), then we might have the information we need. But we would have to somehow tie the debugobject context to the thread context. That sounds too complex for what we are trying to achieve here.
Quoted 3 lines
Indeed! Especially given the fact that RCU read-side critical sections can be nested. Which
rcu_dereference()calls go with which RCU read-side critical section?
Good point.
Quoted 12 lines
Quoted 11 lines
Quoted 4 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.
Quoted 7 lines
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any
rcu_dereference()calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.
Good point too. So something like a new rcu_unreference() (or feel free to find any better name) ;) that would be compiled out normally, but would call into debugobjects might do the trick. We would have to add these annotations to match every rcu_dereference() though, might means a lot of new lines of code. On the plus side, that looks like a good audit of RCU read-side use. ;)
Thanks,
Mathieu
Quoted 30 lines
Thanx, Paul
Quoted 27 lines
Thanks,
Mathieu
Quoted 16 lines
Thanx, Paul
Quoted 13 lines
Thoughts ?
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.comSignature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 4 lines · Mathieu Desnoyers
Quoted 3 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 2 lines · Mathieu Desnoyers
Quoted 1 line · * Paul E. McKenney ([email protected]) wrote:
On Mon, May 17, 2010 at 06:00:25PM -0400, Mathieu Desnoyers wrote:
[ . . . ]
Quoted 19 lines
Quoted 11 lines
Quoted 10 lines
Quoted 3 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.Quoted 7 lines
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any
rcu_dereference()calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.
Quoted 6 lines
Good point too. So something like a new
rcu_unreference()(or feel free to find any better name) ;) that would be compiled out normally, but would call into debugobjects might do the trick. We would have to add these annotations to match everyrcu_dereference()though, might means a lot of new lines of code. On the plus side, that looks like a good audit of RCU read-side use. ;)
My first thought is that we have added quite a bit of RCU consistency check code in the past few months, so we should see what bugs they find and what bugs escape. It is all too easy to create consistency check code that is more trouble than it is worth.
But in the meantime, let's see what would be required to check for failures to insert grace-period delays:
o There would need to be something like rcu_unreference(),
rcu_no_more_readers() or some such after the grace period. The update side would then become something like the following: oldp = rcu_dereference_protected(gp, &mylock); rcu_assign_pointer(gp, newp); synchronize_rcu(); rcu_no_more_readers(oldp); kfree(oldp);
o There would need to be something to check all of the pointers
traversed in the read-side critical sections: rcu_read_lock(); ... p1 = rcu_dereference(gp1->field1); ... p2 = rcu_dereference(gp2->field2); ... rcu_validate(p1); rcu_validate(p2); rcu_read_unlock();
One thing that bothers me about this is that we are forcing the developer to do a lot of extra typing. For example, rcu_no_more_readers() is in a truth-and-beauty sense redundant with kfree() -- why type both? The same could be said about rcu_validate() and rcu_read_unlock(), but nested RCU read-side critical sections make this difficult.
Or am I misunderstanding what you are suggesting?
Thanx, Paul
Michael S. Tsirkin
maintainerRe: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
Quoted 34 lines · Paul E. McKenney
Quoted 4 lines · Mathieu Desnoyers
Quoted 3 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 2 lines · Mathieu Desnoyers
Quoted 1 line · * Paul E. McKenney ([email protected]) wrote:
On Mon, May 17, 2010 at 06:00:25PM -0400, Mathieu Desnoyers wrote:
[ . . . ]
Quoted 19 lines
Quoted 11 lines
Quoted 10 lines
Quoted 3 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.Quoted 7 lines
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any
rcu_dereference()calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.Quoted 6 lines
Good point too. So something like a new
rcu_unreference()(or feel free to find any better name) ;) that would be compiled out normally, but would call into debugobjects might do the trick. We would have to add these annotations to match everyrcu_dereference()though, might means a lot of new lines of code. On the plus side, that looks like a good audit of RCU read-side use. ;)
Quoted 4 lines
My first thought is that we have added quite a bit of RCU consistency check code in the past few months, so we should see what bugs they find and what bugs escape. It is all too easy to create consistency check code that is more trouble than it is worth.
Right. Do the patches that started this discussion catch anything BTW?
Quoted 14 lines
But in the meantime, let's see what would be required to check for failures to insert grace-period delays:
o There would need to be something like
rcu_unreference(),rcu_no_more_readers() or some such after the grace period. The update side would then become something like the following: oldp = rcu_dereference_protected(gp, &mylock); rcu_assign_pointer(gp, newp); synchronize_rcu(); rcu_no_more_readers(oldp); kfree(oldp);o There would need to be something to check all of the pointers
Quoted 13 lines
traversed in the read-side critical sections: rcu_read_lock(); ... p1 = rcu_dereference(gp1->field1); ... p2 = rcu_dereference(gp2->field2); ... rcu_validate(p1); rcu_validate(p2); rcu_read_unlock();
what does rcu_validate do?
Quoted 3 lines
One thing that bothers me about this is that we are forcing the developer to do a lot of extra typing. For example,
rcu_no_more_readers()is in a truth-and-beauty sense redundant withkfree()-- why type both?
With kfree, yes. We could stick rcu_no_more_readers in kfree I guess?
Quoted 6 lines
The
same could be said aboutrcu_validate()andrcu_read_unlock(), but nested RCU read-side critical sections make this difficult.
Or am I misunderstanding what you are suggesting?Thanx, Paul
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Paul E. McKenneyQuoted 34 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 4 lines · Mathieu Desnoyers
Quoted 3 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 2 lines · Mathieu Desnoyers
Quoted 1 line · * Paul E. McKenney ([email protected]) wrote:
On Mon, May 17, 2010 at 06:00:25PM -0400, Mathieu Desnoyers wrote:
[ . . . ]
Quoted 19 lines
Quoted 11 lines
Quoted 10 lines
Quoted 3 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.Quoted 7 lines
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any
rcu_dereference()calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.Quoted 6 lines
Good point too. So something like a new
rcu_unreference()(or feel free to find any better name) ;) that would be compiled out normally, but would call into debugobjects might do the trick. We would have to add these annotations to match everyrcu_dereference()though, might means a lot of new lines of code. On the plus side, that looks like a good audit of RCU read-side use. ;)
Quoted 4 lines
My first thought is that we have added quite a bit of RCU consistency check code in the past few months, so we should see what bugs they find and what bugs escape. It is all too easy to create consistency check code that is more trouble than it is worth.
Yes, although I expect that this new checking scheme will take some time to implement and mainline anyway (implementation effort which I might leave to someone else, as I have to focus on tracing at the moment).
Quoted 4 lines
But in the meantime, let's see what would be required to check for failures to insert grace-period delays:
o There would need to be something like
rcu_unreference(),
Quoted 8 lines
rcu_no_more_readers() or some such after the grace period. The update side would then become something like the following: oldp = rcu_dereference_protected(gp, &mylock); rcu_assign_pointer(gp, newp); synchronize_rcu(); rcu_no_more_readers(oldp); kfree(oldp);
Replacing a kfree with a rcu_free(kfree, oldp) call that would include both could lessen the amount of typing:
#define rcu_free(freefct, ptr) \
do { \
rcu_no_more_readers(ptr); \
freefct(ptr); \
} while (0)
Quoted 12 lines
o There would need to be something to check all of the pointers
traversed in the read-side critical sections: rcu_read_lock(); ... p1 = rcu_dereference(gp1->field1); ... p2 = rcu_dereference(gp2->field2); ... rcu_validate(p1); rcu_validate(p2);
Hrm, isn't the goal of this "rcu_validate(p1)" just to keep track of "p1" liveness ? Or do you plan to add a check there also ? I'm not sure I figure out what you are planning to validate here. I was thinking more in terms of
rcu_unreference(p1);
rcu_unreference(p1);that would be symmetric with the rcu_dereference.
Quoted 7 lines
rcu_read_unlock();One thing that bothers me about this is that we are forcing the developer to do a lot of extra typing. For example,
rcu_no_more_readers()is in a truth-and-beauty sense redundant withkfree()-- why type both? The same could be said aboutrcu_validate()andrcu_read_unlock(), but nested RCU read-side critical sections make this difficult.
Ideally we'd like to add near-zero burden on developers, but I fear this cannot be done easily for read-side C.S.. As for write-side, we have to choose between tradeoff of genericity and less typing, e.g., between:
rcu_free(kfree, ptr);
and
rcu_kfree(ptr)
for the second, we would have to create a whole family of rcu_*free().
Quoted 2 lines
Or am I misunderstanding what you are suggesting?
I'm only unsure about the "validate" part.
Thanks,
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Michael S. TsirkinQuoted 35 lines · Michael S. Tsirkin
Quoted 34 lines · Paul E. McKenney
Quoted 4 lines · Mathieu Desnoyers
Quoted 3 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 2 lines · Mathieu Desnoyers
Quoted 1 line · * Paul E. McKenney ([email protected]) wrote:
On Mon, May 17, 2010 at 06:00:25PM -0400, Mathieu Desnoyers wrote:
[ . . . ]
Quoted 25 lines
Quoted 11 lines
Quoted 10 lines
Quoted 3 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.Quoted 7 lines
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any
rcu_dereference()calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.Good point too. So something like a new
rcu_unreference()(or feel free to find any better name) ;) that would be compiled out normally, but would call into debugobjects might do the trick. We would have to add these annotations to match everyrcu_dereference()though, might means a lot of new lines of code. On the plus side, that looks like a good audit of RCU read-side use. ;)
Quoted 6 lines
Quoted 4 lines
My first thought is that we have added quite a bit of RCU consistency check code in the past few months, so we should see what bugs they find and what bugs escape. It is all too easy to create consistency check code that is more trouble than it is worth.
Right. Do the patches that started this discussion catch anything BTW?
All three approaches have found some bugs.
Quoted 23 lines
Quoted 23 lines
But in the meantime, let's see what would be required to check for failures to insert grace-period delays:
o There would need to be something like
rcu_unreference(),rcu_no_more_readers() or some such after the grace period. The update side would then become something like the following: oldp = rcu_dereference_protected(gp, &mylock); rcu_assign_pointer(gp, newp); synchronize_rcu(); rcu_no_more_readers(oldp); kfree(oldp);o There would need to be something to check all of the pointers
traversed in the read-side critical sections: rcu_read_lock(); ... p1 = rcu_dereference(gp1->field1); ... p2 = rcu_dereference(gp2->field2); ...
Quoted 6 lines
Quoted 4 lines
rcu_validate(p1); rcu_validate(p2); rcu_read_unlock();what does
rcu_validatedo?
It checks to make sure that the pointer still points to something valid.
Quoted 5 lines
Quoted 3 lines
One thing that bothers me about this is that we are forcing the developer to do a lot of extra typing. For example,
rcu_no_more_readers()is in a truth-and-beauty sense redundant withkfree()-- why type both?With kfree, yes. We could stick
rcu_no_more_readersin kfree I guess?
But why not just use the existing debugobjects? You can just use something like this:
debug_check_no_obj_freed(p1, sizeof(*p1));
in place of:
rcu_validate(p1);
Of course, if you are using your own custom allocator, you will need to put the allocation/free checks in, same as slab and the others currently do.
Thanx, Paul
Quoted 6 lines
Quoted 6 lines
The
same could be said aboutrcu_validate()andrcu_read_unlock(), but nested RCU read-side critical sections make this difficult.
Or am I misunderstanding what you are suggesting?Thanx, Paul
Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations
In reply to Mathieu DesnoyersQuoted 57 lines · Mathieu Desnoyers
Quoted 34 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 4 lines · Mathieu Desnoyers
Quoted 3 lines · * Paul E. McKenney ([email protected]) wrote:
Quoted 2 lines · Mathieu Desnoyers
Quoted 1 line · * Paul E. McKenney ([email protected]) wrote:
On Mon, May 17, 2010 at 06:00:25PM -0400, Mathieu Desnoyers wrote:
[ . . . ]
Quoted 19 lines
Quoted 11 lines
Quoted 10 lines
Quoted 3 lines
But perhaps we should be simply treating this as a use-after-free problem, so that RCU is not directly involved. Isn't that the standard use of debugobjects anyway?
OK so we could tie "
rcu_dereference" do debugobjects, and free would be a standard free. Yes, I think it could be done. It looks a bit like the memory allocation debugging code. If we know that a certainrcu_dereferencealways access dynamically allocated memory, we could probably add some checks there based on the memory allocator debug objects.Quoted 7 lines
We probably need vhost to add code at the end of the relevant RCU read-side critical section checking that the pointers returned by any
rcu_dereference()calls still point to valid memory. Don't get me wrong, your approach could find bugs in which someone forgot to remove the RCU-protected structure from a public list, but it could not detect failure to wait a grace period between the time of removal and the time of freeing.Quoted 6 lines
Good point too. So something like a new
rcu_unreference()(or feel free to find any better name) ;) that would be compiled out normally, but would call into debugobjects might do the trick. We would have to add these annotations to match everyrcu_dereference()though, might means a lot of new lines of code. On the plus side, that looks like a good audit of RCU read-side use. ;)Quoted 4 lines
My first thought is that we have added quite a bit of RCU consistency check code in the past few months, so we should see what bugs they find and what bugs escape. It is all too easy to create consistency check code that is more trouble than it is worth.
Yes, although I expect that this new checking scheme will take some time to implement and mainline anyway (implementation effort which I might leave to someone else, as I have to focus on tracing at the moment).
Quoted 4 lines
But in the meantime, let's see what would be required to check for failures to insert grace-period delays:
o There would need to be something like
rcu_unreference(),Quoted 8 lines
rcu_no_more_readers() or some such after the grace period. The update side would then become something like the following: oldp = rcu_dereference_protected(gp, &mylock); rcu_assign_pointer(gp, newp); synchronize_rcu(); rcu_no_more_readers(oldp); kfree(oldp);
Quoted 8 lines
Replacing a kfree with a
rcu_free(kfree, oldp)call that would include both could lessen the amount of typing:#define
rcu_free(freefct, ptr)\
do { \
rcu_no_more_readers(ptr); \
freefct(ptr); \
} while (0)
Or we could just rely on the existing debugobjects support that is already in kfree(). ;-)
Quoted 18 lines
Quoted 12 lines
o There would need to be something to check all of the pointers
traversed in the read-side critical sections: rcu_read_lock(); ... p1 = rcu_dereference(gp1->field1); ... p2 = rcu_dereference(gp2->field2); ... rcu_validate(p1); rcu_validate(p2);Hrm, isn't the goal of this "
rcu_validate(p1)" just to keep track of "p1" liveness ? Or do you plan to add a check there also ? I'm not sure I figure out what you are planning to validate here. I was thinking more in terms of
Quoted 4 lines
rcu_unreference(p1); rcu_unreference(p1);that would be symmetric with the
rcu_dereference.
My preference would be for people to just use the existing debugobjects API, debug_check_no_obj_freed(). That is already in place, no need to create RCU wrappers for it.
Quoted 16 lines
Quoted 7 lines
rcu_read_unlock();One thing that bothers me about this is that we are forcing the developer to do a lot of extra typing. For example,
rcu_no_more_readers()is in a truth-and-beauty sense redundant withkfree()-- why type both? The same could be said aboutrcu_validate()andrcu_read_unlock(), but nested RCU read-side critical sections make this difficult.Ideally we'd like to add near-zero burden on developers, but I fear this cannot be done easily for read-side
C.S.. As for write-side, we have to choose between tradeoff of genericity and less typing, e.g., between:
rcu_free(kfree, ptr);
and
rcu_kfree(ptr)
Quoted 6 lines
for the second, we would have to create a whole family of rcu_*free().
Quoted 2 lines
Or am I misunderstanding what you are suggesting?
I'm only unsure about the "validate" part.
Again, we should just rely on the existing debugobjects function, letting developers use it as they see fit.
Thanx, Paul
Quoted 12 lines
Thanks,
Mathieu
Quoted 2 lines
Thanx, Paul
Signature · 4 lines
Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com