virtual-bus

new file mode 100644
index 0000000..a65e54e
@@ -0,0 +1,188 @@
+/*
+ * Copyright 2009 Novell.  All Rights Reserved.
+ *
+ * Author:
+ *      Gregory Haskins <[email protected]>
+ *
+ * This file is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#ifndef _LINUX_SHM_SIGNAL_H
+#define _LINUX_SHM_SIGNAL_H
+
+#include <asm/types.h>
+
+/*
+ *---------
+ * The following structures represent data that is shared across boundaries
+ * which may be quite disparate from one another (e.g. Windows vs Linux,
+ * 32 vs 64 bit, etc).  Therefore, care has been taken to make sure they
+ * present data in a manner that is independent of the environment.
+ *-----------
+ */
+
+#define SHM_SIGNAL_MAGIC 0x58fa39df
+#define SHM_SIGNAL_VER   1
+
+struct shm_signal_irq {
+	__u8                  enabled;
+	__u8                  pending;
+	__u8                  dirty;
+};
+
+enum shm_signal_locality {
+	shm_locality_north,
+	shm_locality_south,
+};
+
+struct shm_signal_desc {
+	__u32                 magic;
+	__u32                 ver;
+	struct shm_signal_irq irq[2];
+};
+
+/* --- END SHARED STRUCTURES --- */
+
+#ifdef __KERNEL__
+
+#include <linux/interrupt.h>
+
+struct shm_signal_notifier {
+	void (*signal)(struct shm_signal_notifier *);
+};
+
+struct shm_signal;
+
+struct shm_signal_ops {
+	int      (*inject)(struct shm_signal *s);
+	void     (*fault)(struct shm_signal *s, const char *fmt, ...);
+	void     (*release)(struct shm_signal *s);
+};
+
+enum {
+	shm_signal_in_wakeup,
+};
+
+struct shm_signal {
+	atomic_t                    refs;
+	spinlock_t                  lock;
+	enum shm_signal_locality    locale;
+	unsigned long               flags;
+	struct shm_signal_ops      *ops;
+	struct shm_signal_desc     *desc;
+	struct shm_signal_notifier *notifier;
+	struct tasklet_struct       deferred_notify;
+};
+
+#define SHM_SIGNAL_FAULT(s, fmt, args...)  \
+  ((s)->ops->fault ? (s)->ops->fault((s), fmt, ## args) : panic(fmt, ## args))
+
+ /*
+  * These functions should only be used internally
+  */
+void _shm_signal_release(struct shm_signal *s);
+void _shm_signal_wakeup(struct shm_signal *s);
+
+/**
+ * shm_signal_init() - initialize an SHM_SIGNAL
+ * @s:        SHM_SIGNAL context
+ *
+ * Initializes SHM_SIGNAL context before first use
+ *
+ **/
+void shm_signal_init(struct shm_signal *s);
+
+/**
+ * shm_signal_get() - acquire an SHM_SIGNAL context reference
+ * @s:        SHM_SIGNAL context
+ *
+ **/
+static inline struct shm_signal *shm_signal_get(struct shm_signal *s)
+{
+	atomic_inc(&s->refs);
+
+	return s;
+}
+
+/**
+ * shm_signal_put() - release an SHM_SIGNAL context reference
+ * @s:        SHM_SIGNAL context
+ *
+ **/
+static inline void shm_signal_put(struct shm_signal *s)
+{
+	if (atomic_dec_and_test(&s->refs))
+		_shm_signal_release(s);
+}
+
+/**
+ * shm_signal_enable() - enables local notifications on an SHM_SIGNAL
+ * @s:        SHM_SIGNAL context
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Enables/unmasks the registered notifier (if applicable) to receive wakeups
+ * whenever the remote side performs an shm_signal() operation. A notification
+ * will be dispatched immediately if any pending signals have already been
+ * issued prior to invoking this call.
+ *
+ * This is synonymous with unmasking an interrupt.
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int shm_signal_enable(struct shm_signal *s, int flags);
+
+/**
+ * shm_signal_disable() - disable local notifications on an SHM_SIGNAL
+ * @s:        SHM_SIGNAL context
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Disables/masks the registered shm_signal_notifier (if applicable) from
+ * receiving any further notifications.  Any subsequent calls to shm_signal()
+ * by the remote side will update the shm as dirty, but will not traverse the
+ * locale boundary and will not invoke the notifier callback.  Signals
+ * delivered while masked will be deferred until shm_signal_enable() is
+ * invoked.
+ *
+ * This is synonymous with masking an interrupt
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int shm_signal_disable(struct shm_signal *s, int flags);
+
+/**
+ * shm_signal_inject() - notify the remote side about shm changes
+ * @s:        SHM_SIGNAL context
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Marks the shm state as "dirty" and, if enabled, will traverse
+ * a locale boundary to inject a remote notification.  The remote
+ * side controls whether the notification should be delivered via
+ * the shm_signal_enable/disable() interface.
+ *
+ * The specifics of how to traverse a locale boundary are abstracted
+ * by the shm_signal_ops->signal() interface and provided by a particular
+ * implementation.  However, typically going north to south would be
+ * something like a syscall/hypercall, and going south to north would be
+ * something like a posix-signal/guest-interrupt.
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int shm_signal_inject(struct shm_signal *s, int flags);
+
+#endif /* __KERNEL__ */
+
+#endif /* _LINUX_SHM_SIGNAL_H */

186 lines not on this page · raw message

114 lines not on this page · raw message

index 011db2f..cd2f9b1 100644
@@ -97,6 +97,7 @@ struct futex_pi_state;
 struct robust_list_head;
 struct bio;
 struct bts_tracer;
+struct vbus;
 
 /*
  * List of flags we want to share for kernel threads,

@@ -1329,6 +1330,9 @@ struct task_struct {
 	unsigned int lockdep_recursion;
 	struct held_lock held_locks[MAX_LOCK_DEPTH];
 #endif
+#ifdef CONFIG_VBUS
+	struct vbus *vbus;
+#endif
 
 /* journalling filesystem info */
 	void *journal_info;

147 lines not on this page · raw message

417 lines not on this page · raw message

index efd30cc..8736de6 100644
@@ -48,6 +48,7 @@
 #include <linux/tracehook.h>
 #include <linux/init_task.h>
 #include <trace/sched.h>
+#include <linux/vbus.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>

@@ -1081,6 +1082,7 @@ NORET_TYPE void do_exit(long code)
 	check_stack_usage();
 	exit_thread();
 	cgroup_exit(tsk, 1);
+	task_vbus_disassociate(tsk);
 
 	if (group_dead && tsk->signal->leader)
 		disassociate_ctty(1);

index 4854c2c..5536053 100644
@@ -61,6 +61,7 @@
 #include <linux/proc_fs.h>
 #include <linux/blkdev.h>
 #include <trace/sched.h>
+#include <linux/vbus.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>

@@ -1274,6 +1275,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	write_unlock_irq(&tasklist_lock);
 	proc_fork_connector(p);
 	cgroup_post_fork(p);
+	fork_vbus(p);
 	return p;
 
 bad_fork_free_graph:

52 lines not on this page · raw message

275 lines not on this page · raw message

567 lines not on this page · raw message

124 lines not on this page · raw message

72 lines not on this page · raw message

41 lines not on this page · raw message

116 lines not on this page · raw message

new file mode 100644
index 0000000..62dab78
@@ -0,0 +1,115 @@
+/*
+ * Copyright 2009 Novell.  All Rights Reserved.
+ *
+ * Virtual-Bus - Client interface
+ *
+ * We expect to have various types of connection-clients (e.g. userspace,
+ * kvm, etc).  Each client will be connecting from some environment outside
+ * of the kernel, and therefore will not have direct access to the API as
+ * presented in ./linux/vbus.h.  There will undoubtedly be some parameter
+ * marshalling that must occur, as well as common patterns for the handling
+ * of those marshalled parameters (e.g. translating a handle into a pointer,
+ * etc).
+ *
+ * Therefore this "client" API is provided to simplify the development
+ * of any clients.  Of course, a client is free to bypass this API entirely
+ * and communicate with the direct VBUS API if desired.
+ *
+ * Author:
+ *      Gregory Haskins <[email protected]>
+ *
+ * This file is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#ifndef _LINUX_VBUS_CLIENT_H
+#define _LINUX_VBUS_CLIENT_H
+
+#include <linux/types.h>
+#include <linux/compiler.h>
+
+struct vbus_deviceopen {
+	__u32 devid;
+	__u32 version; /* device ABI version */
+	__u64 handle; /* return value for devh */
+};
+
+struct vbus_devicecall {
+	__u64 devh;   /* device-handle (returned from DEVICEOPEN */
+	__u32 func;
+	__u32 len;
+	__u32 flags;
+	__u64 datap;
+};
+
+struct vbus_deviceshm {
+	__u64 devh;   /* device-handle (returned from DEVICEOPEN */
+	__u32 id;
+	__u32 len;
+	__u32 flags;
+	struct {
+		__u32 offset;
+		__u32 prio;
+		__u64 cookie; /* token to pass back when signaling client */
+	} signal;
+	__u64 datap;
+	__u64 handle; /* return value for signaling from client to kernel */
+};
+
+#ifdef __KERNEL__
+
+#include <linux/ioq.h>
+#include <linux/module.h>
+#include <asm/atomic.h>
+
+struct vbus_client;
+
+struct vbus_client_ops {
+	int (*deviceopen)(struct vbus_client *client,  struct vbus_memctx *ctx,
+			  __u32 devid, __u32 version, __u64 *devh);
+	int (*deviceclose)(struct vbus_client *client, __u64 devh);
+	int (*devicecall)(struct vbus_client *client,
+			  __u64 devh, __u32 func,
+			  void *data, __u32 len, __u32 flags);
+	int (*deviceshm)(struct vbus_client *client,
+			 __u64 devh, __u32 id,
+			 struct vbus_shm *shm, struct shm_signal *signal,
+			 __u32 flags, __u64 *handle);
+	int (*shmsignal)(struct vbus_client *client, __u64 handle);
+	void (*release)(struct vbus_client *client);
+};
+
+struct vbus_client {
+	atomic_t refs;
+	struct vbus_client_ops *ops;
+};
+
+static inline void vbus_client_get(struct vbus_client *client)
+{
+	atomic_inc(&client->refs);
+}
+
+static inline void vbus_client_put(struct vbus_client *client)
+{
+	if (atomic_dec_and_test(&client->refs))
+		client->ops->release(client);
+}
+
+struct vbus_client *vbus_client_attach(struct vbus *bus);
+
+extern struct vbus_memctx *current_memctx;
+struct vbus_memctx *task_memctx_alloc(struct task_struct *task);
+
+#endif /* __KERNEL__ */
+
+#endif /* _LINUX_VBUS_CLIENT_H */

543 lines not on this page · raw message

index 5f0566c..04db4ff 100644
@@ -29,6 +29,7 @@
 #include <linux/sched.h>
 #include <linux/rcupdate.h>
 #include <linux/vbus_device.h>
+#include <linux/notifier.h>
 
 struct vbus;
 struct task_struct;

@@ -137,6 +138,20 @@ static inline void task_vbus_disassociate(struct task_struct *p)
 	}
 }
 
+enum {
+	VBUS_EVENT_DEVADD,
+	VBUS_EVENT_DEVDROP,
+};
+
+struct vbus_event_devadd {
+	const char   *type;
+	unsigned long id;
+};
+
+int vbus_notifier_register(struct vbus *vbus, struct notifier_block *nb);
+int vbus_notifier_unregister(struct vbus *vbus, struct notifier_block *nb);
+
+
 #else /* CONFIG_VBUS */
 
 #define fork_vbus(p) do { } while (0)

89 lines not on this page · raw message

index 1266d69..cd2676b 100644
@@ -51,6 +51,7 @@ struct vbus {
 	struct vbus_subdir members;
 	unsigned long next_id;
 	struct rb_node node;
+	struct raw_notifier_head notifier;
 };
 
 struct vbus_member {

new file mode 100644
index 0000000..c53e13f
@@ -0,0 +1,73 @@
+/*
+ * Copyright 2009 Novell.  All Rights Reserved.
+ *
+ * Mediates access to a host VBUS from a guest kernel by providing a
+ * global view of all VBUS devices
+ *
+ * Author:
+ *      Gregory Haskins <[email protected]>
+ *
+ * This file is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#ifndef _LINUX_VBUS_DRIVER_H
+#define _LINUX_VBUS_DRIVER_H
+
+#include <linux/device.h>
+#include <linux/shm_signal.h>
+
+struct vbus_device_proxy;
+struct vbus_driver;
+
+struct vbus_device_proxy_ops {
+	int (*open)(struct vbus_device_proxy *dev, int version, int flags);
+	int (*close)(struct vbus_device_proxy *dev, int flags);
+	int (*shm)(struct vbus_device_proxy *dev, int id, int prio,
+		   void *ptr, size_t len,
+		   struct shm_signal_desc *sigdesc, struct shm_signal **signal,
+		   int flags);
+	int (*call)(struct vbus_device_proxy *dev, u32 func,
+		    void *data, size_t len, int flags);
+	void (*release)(struct vbus_device_proxy *dev);
+};
+
+struct vbus_device_proxy {
+	char                          *type;
+	u64                            id;
+	void                          *priv; /* Used by drivers */
+	struct vbus_device_proxy_ops  *ops;
+	struct device                  dev;
+};
+
+int vbus_device_proxy_register(struct vbus_device_proxy *dev);
+void vbus_device_proxy_unregister(struct vbus_device_proxy *dev);
+
+struct vbus_device_proxy *vbus_device_proxy_find(u64 id);
+
+struct vbus_driver_ops {
+	int (*probe)(struct vbus_device_proxy *dev);
+	int (*remove)(struct vbus_device_proxy *dev);
+};
+
+struct vbus_driver {
+	char                          *type;
+	struct module                 *owner;
+	struct vbus_driver_ops        *ops;
+	struct device_driver           drv;
+};
+
+int vbus_driver_register(struct vbus_driver *drv);
+void vbus_driver_unregister(struct vbus_driver *drv);
+
+#endif /* _LINUX_VBUS_DRIVER_H */

152 lines not on this page · raw message

new file mode 100644
index 0000000..d450d9a
@@ -0,0 +1,410 @@
+/*
+ * Copyright 2009 Novell.  All Rights Reserved.
+ *
+ * IOQ is a generic shared-memory, lockless queue mechanism. It can be used
+ * in a variety of ways, though its intended purpose is to become the
+ * asynchronous communication path for virtual-bus drivers.
+ *
+ * The following are a list of key design points:
+ *
+ * #) All shared-memory is always allocated on explicitly one side of the
+ *    link.  This typically would be the guest side in a VM/VMM scenario.
+ * #) Each IOQ has the concept of "north" and "south" locales, where
+ *    north denotes the memory-owner side (e.g. guest).
+ * #) An IOQ is manipulated using an iterator idiom.
+ * #) Provides a bi-directional signaling/notification infrastructure on
+ *    a per-queue basis, which includes an event mitigation strategy
+ *    to reduce boundary switching.
+ * #) The signaling path is abstracted so that various technologies and
+ *    topologies can define their own specific implementation while sharing
+ *    the basic structures and code.
+ *
+ * Author:
+ *      Gregory Haskins <[email protected]>
+ *
+ * This file is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#ifndef _LINUX_IOQ_H
+#define _LINUX_IOQ_H
+
+#include <asm/types.h>
+#include <linux/shm_signal.h>
+
+/*
+ *---------
+ * The following structures represent data that is shared across boundaries
+ * which may be quite disparate from one another (e.g. Windows vs Linux,
+ * 32 vs 64 bit, etc).  Therefore, care has been taken to make sure they
+ * present data in a manner that is independent of the environment.
+ *-----------
+ */
+struct ioq_ring_desc {
+	__u64                 cookie; /* for arbitrary use by north-side */
+	__u64                 ptr;
+	__u64                 len;
+	__u8                  valid;
+	__u8                  sown; /* South owned = 1, North owned = 0 */
+};
+
+#define IOQ_RING_MAGIC 0x47fa2fe4
+#define IOQ_RING_VER   4
+
+struct ioq_ring_idx {
+	__u32                 head;    /* 0 based index to head of ptr array */
+	__u32                 tail;    /* 0 based index to tail of ptr array */
+	__u8                  full;
+};
+
+enum ioq_locality {
+	ioq_locality_north,
+	ioq_locality_south,
+};
+
+struct ioq_ring_head {
+	__u32                  magic;
+	__u32                  ver;
+	struct shm_signal_desc signal;
+	struct ioq_ring_idx    idx[2];
+	__u32                  count;
+	struct ioq_ring_desc   ring[1]; /* "count" elements will be allocated */
+};
+
+#define IOQ_HEAD_DESC_SIZE(count) \
+    (sizeof(struct ioq_ring_head) + sizeof(struct ioq_ring_desc) * (count - 1))
+
+/* --- END SHARED STRUCTURES --- */
+
+#ifdef __KERNEL__
+
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/interrupt.h>
+#include <linux/shm_signal.h>
+#include <asm/atomic.h>
+
+enum ioq_idx_type {
+	ioq_idxtype_valid,
+	ioq_idxtype_inuse,
+	ioq_idxtype_both,
+	ioq_idxtype_invalid,
+};
+
+enum ioq_seek_type {
+	ioq_seek_tail,
+	ioq_seek_next,
+	ioq_seek_head,
+	ioq_seek_set
+};
+
+struct ioq_iterator {
+	struct ioq            *ioq;
+	struct ioq_ring_idx   *idx;
+	u32                    pos;
+	struct ioq_ring_desc  *desc;
+	int                    update:1;
+	int                    dualidx:1;
+	int                    flipowner:1;
+};
+
+struct ioq_notifier {
+	void (*signal)(struct ioq_notifier *);
+};
+
+struct ioq_ops {
+	void     (*release)(struct ioq *ioq);
+};
+
+struct ioq {
+	struct ioq_ops *ops;
+
+	atomic_t               refs;
+	enum ioq_locality      locale;
+	struct ioq_ring_head  *head_desc;
+	struct ioq_ring_desc  *ring;
+	struct shm_signal     *signal;
+	wait_queue_head_t      wq;
+	struct ioq_notifier   *notifier;
+	size_t                 count;
+	struct shm_signal_notifier shm_notifier;
+};
+
+#define IOQ_ITER_AUTOUPDATE  (1 << 0)
+#define IOQ_ITER_NOFLIPOWNER (1 << 1)
+
+/**
+ * ioq_init() - initialize an IOQ
+ * @ioq:        IOQ context
+ *
+ * Initializes IOQ context before first use
+ *
+ **/
+void ioq_init(struct ioq *ioq,
+	      struct ioq_ops *ops,
+	      enum ioq_locality locale,
+	      struct ioq_ring_head *head,
+	      struct shm_signal *signal,
+	      size_t count);
+
+/**
+ * ioq_get() - acquire an IOQ context reference
+ * @ioq:        IOQ context
+ *
+ **/
+static inline struct ioq *ioq_get(struct ioq *ioq)
+{
+	atomic_inc(&ioq->refs);
+
+	return ioq;
+}
+
+/**
+ * ioq_put() - release an IOQ context reference
+ * @ioq:        IOQ context
+ *
+ **/
+static inline void ioq_put(struct ioq *ioq)
+{
+	if (atomic_dec_and_test(&ioq->refs)) {
+		shm_signal_put(ioq->signal);
+		ioq->ops->release(ioq);
+	}
+}
+
+/**
+ * ioq_notify_enable() - enables local notifications on an IOQ
+ * @ioq:        IOQ context
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Enables/unmasks the registered ioq_notifier (if applicable) and waitq to
+ * receive wakeups whenever the remote side performs an ioq_signal() operation.
+ * A notification will be dispatched immediately if any pending signals have
+ * already been issued prior to invoking this call.
+ *
+ * This is synonymous with unmasking an interrupt.
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+static inline int ioq_notify_enable(struct ioq *ioq, int flags)
+{
+	return shm_signal_enable(ioq->signal, 0);
+}
+
+/**
+ * ioq_notify_disable() - disable local notifications on an IOQ
+ * @ioq:        IOQ context
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Disables/masks the registered ioq_notifier (if applicable) and waitq
+ * from receiving any further notifications.  Any subsequent calls to
+ * ioq_signal() by the remote side will update the ring as dirty, but
+ * will not traverse the locale boundary and will not invoke the notifier
+ * callback or wakeup the waitq.  Signals delivered while masked will
+ * be deferred until ioq_notify_enable() is invoked
+ *
+ * This is synonymous with masking an interrupt
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+static inline int ioq_notify_disable(struct ioq *ioq, int flags)
+{
+	return shm_signal_disable(ioq->signal, 0);
+}
+
+/**
+ * ioq_signal() - notify the remote side about ring changes
+ * @ioq:        IOQ context
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Marks the ring state as "dirty" and, if enabled, will traverse
+ * a locale boundary to invoke a remote notification.  The remote
+ * side controls whether the notification should be delivered via
+ * the ioq_notify_enable/disable() interface.
+ *
+ * The specifics of how to traverse a locale boundary are abstracted
+ * by the ioq_ops->signal() interface and provided by a particular
+ * implementation.  However, typically going north to south would be
+ * something like a syscall/hypercall, and going south to north would be
+ * something like a posix-signal/guest-interrupt.
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+static inline int ioq_signal(struct ioq *ioq, int flags)
+{
+	return shm_signal_inject(ioq->signal, 0);
+}
+
+/**
+ * ioq_count() - counts the number of outstanding descriptors in an index
+ * @ioq:        IOQ context
+ * @type:	Specifies the index type
+ *                 (*) valid: the descriptor is valid.  This is usually
+ *                     used to keep track of descriptors that may not
+ *                     be carrying a useful payload, but still need to
+ *                     be tracked carefully.
+ *                 (*) inuse: Descriptors that carry useful payload
+ *
+ * Returns:
+ *  (*) >=0: # of descriptors outstanding in the index
+ *  (*) <0 = ERRNO
+ *
+ **/
+int ioq_count(struct ioq *ioq, enum ioq_idx_type type);
+
+/**
+ * ioq_remain() - counts the number of remaining descriptors in an index
+ * @ioq:        IOQ context
+ * @type:	Specifies the index type
+ *                 (*) valid: the descriptor is valid.  This is usually
+ *                     used to keep track of descriptors that may not
+ *                     be carrying a useful payload, but still need to
+ *                     be tracked carefully.
+ *                 (*) inuse: Descriptors that carry useful payload
+ *
+ * This is the converse of ioq_count().  This function returns the number
+ * of "free" descriptors left in a particular index
+ *
+ * Returns:
+ *  (*) >=0: # of descriptors remaining in the index
+ *  (*) <0 = ERRNO
+ *
+ **/
+int ioq_remain(struct ioq *ioq, enum ioq_idx_type type);
+
+/**
+ * ioq_size() - counts the maximum number of descriptors in an ring
+ * @ioq:        IOQ context
+ *
+ * This function returns the maximum number of descriptors supported in
+ * a ring, regardless of their current state (free or inuse).
+ *
+ * Returns:
+ *  (*) >=0: total # of descriptors in the ring
+ *  (*) <0 = ERRNO
+ *
+ **/
+int ioq_size(struct ioq *ioq);
+
+/**
+ * ioq_full() - determines if a specific index is "full"
+ * @ioq:        IOQ context
+ * @type:	Specifies the index type
+ *                 (*) valid: the descriptor is valid.  This is usually
+ *                     used to keep track of descriptors that may not
+ *                     be carrying a useful payload, but still need to
+ *                     be tracked carefully.
+ *                 (*) inuse: Descriptors that carry useful payload
+ *
+ * Returns:
+ *  (*) 0: index is not full
+ *  (*) 1: index is full
+ *  (*) <0 = ERRNO
+ *
+ **/
+int ioq_full(struct ioq *ioq, enum ioq_idx_type type);
+
+/**
+ * ioq_empty() - determines if a specific index is "empty"
+ * @ioq:        IOQ context
+ * @type:	Specifies the index type
+ *                 (*) valid: the descriptor is valid.  This is usually
+ *                     used to keep track of descriptors that may not
+ *                     be carrying a useful payload, but still need to
+ *                     be tracked carefully.
+ *                 (*) inuse: Descriptors that carry useful payload
+ *
+ * Returns:
+ *  (*) 0: index is not empty
+ *  (*) 1: index is empty
+ *  (*) <0 = ERRNO
+ *
+ **/
+static inline int ioq_empty(struct ioq *ioq, enum ioq_idx_type type)
+{
+    return !ioq_count(ioq, type);
+}
+
+/**
+ * ioq_iter_init() - initialize an iterator for IOQ descriptor traversal
+ * @ioq:        IOQ context to iterate on
+ * @iter:	Iterator context to init (usually from stack)
+ * @type:	Specifies the index type to iterate against
+ *                 (*) valid: iterate against the "valid" index
+ *                 (*) inuse: iterate against the "inuse" index
+ *                 (*) both: iterate against both indexes simultaneously
+ * @flags:      Bitfield with 0 or more bits set to alter behavior
+ *                 (*) autoupdate: automatically signal the remote side
+ *                     whenever the iterator pushes/pops to a new desc
+ *                 (*) noflipowner: do not flip the ownership bit during
+ *                     a push/pop operation
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int ioq_iter_init(struct ioq *ioq, struct ioq_iterator *iter,
+		  enum ioq_idx_type type, int flags);
+
+/**
+ * ioq_iter_seek() - seek to a specific location in the IOQ ring
+ * @iter:	Iterator context (must be initialized with ioq_iter_init)
+ * @type:	Specifies the type of seek operation
+ *                 (*) tail: seek to the absolute tail, offset is ignored
+ *                 (*) next: seek to the relative next, offset is ignored
+ *                 (*) head: seek to the absolute head, offset is ignored
+ *                 (*) set: seek to the absolute offset
+ * @offset:     Offset for ioq_seek_set operations
+ * @flags:      Reserved for future use, must be 0
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int  ioq_iter_seek(struct ioq_iterator *iter, enum ioq_seek_type type,
+		   long offset, int flags);
+
+/**
+ * ioq_iter_push() - push the tail pointer forward
+ * @iter:	Iterator context (must be initialized with ioq_iter_init)
+ * @flags:      Reserved for future use, must be 0
+ *
+ * This function will simultaneously advance the tail ptr in the current
+ * index (valid/inuse, as specified in the ioq_iter_init) as well as
+ * perform a seek(next) operation.  This effectively "pushes" a new pointer
+ * onto the tail of the index.
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int  ioq_iter_push(struct ioq_iterator *iter, int flags);
+
+/**
+ * ioq_iter_pop() - pop the head pointer from the ring
+ * @iter:	Iterator context (must be initialized with ioq_iter_init)
+ * @flags:      Reserved for future use, must be 0
+ *
+ * This function will simultaneously advance the head ptr in the current
+ * index (valid/inuse, as specified in the ioq_iter_init) as well as
+ * perform a seek(next) operation.  This effectively "pops" a pointer
+ * from the head of the index.
+ *
+ * Returns: success = 0, <0 = ERRNO
+ *
+ **/
+int  ioq_iter_pop(struct ioq_iterator *iter,  int flags);
+
+#endif /* __KERNEL__ */
+
+#endif /* _LINUX_IOQ_H */

298 lines not on this page · raw message

index f73cd86..c91bce4 100644
@@ -102,6 +102,7 @@
 #include <linux/configfs.h>
 #include <linux/rbtree.h>
 #include <linux/shm_signal.h>
+#include <linux/ioq.h>
 #include <linux/vbus.h>
 #include <asm/atomic.h>
 

@@ -414,4 +415,10 @@ static inline void vbus_connection_put(struct vbus_connection *conn)
 		conn->ops->release(conn);
 }
 
+/*
+ * device-side IOQ helper - dereferences device-shm as an IOQ
+ */
+int vbus_shm_ioq_attach(struct vbus_shm *shm, struct shm_signal *signal,
+			int maxcount, struct ioq **ioq);
+
 #endif /* _LINUX_VBUS_DEVICE_H */

index c53e13f..9cfbf60 100644
@@ -26,6 +26,7 @@
 
 #include <linux/device.h>
 #include <linux/shm_signal.h>
+#include <linux/ioq.h>
 
 struct vbus_device_proxy;
 struct vbus_driver;

@@ -70,4 +71,10 @@ struct vbus_driver {
 int vbus_driver_register(struct vbus_driver *drv);
 void vbus_driver_unregister(struct vbus_driver *drv);
 
+/*
+ * driver-side IOQ helper - allocates device-shm and maps an IOQ on it
+ */
+int vbus_driver_ioq_alloc(struct vbus_device_proxy *dev, int id, int prio,
+			  size_t ringsize, struct ioq **ioq);
+
 #endif /* _LINUX_VBUS_DRIVER_H */

67 lines not on this page · raw message

89 lines not on this page · raw message

new file mode 100644
index 0000000..ef6b199
@@ -0,0 +1,47 @@
+/*
+ * Copyright 2008 Novell.  All Rights Reserved.
+ *
+ * Virtual-Ethernet adapter
+ *
+ * Author:
+ *      Gregory Haskins <[email protected]>
+ *
+ * This file is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
+ */
+
+#ifndef _LINUX_VENET_H
+#define _LINUX_VENET_H
+
+#define VENET_VERSION 1
+
+#define VENET_TYPE "virtual-ethernet"
+
+#define VENET_QUEUE_RX 0
+#define VENET_QUEUE_TX 1
+
+struct venet_capabilities {
+	__u32 gid;
+	__u32 bits;
+};
+
+/* CAPABILITIES-GROUP 0 */
+/* #define VENET_CAP_FOO    0   (No capabilities defined yet, for now) */
+
+#define VENET_FUNC_LINKUP   0
+#define VENET_FUNC_LINKDOWN 1
+#define VENET_FUNC_MACQUERY 2
+#define VENET_FUNC_NEGCAP   3 /* negotiate capabilities */
+#define VENET_FUNC_FLUSHRX  4
+
+#endif /* _LINUX_VENET_H */

680 lines not on this page · raw message

1,388 lines not on this page · raw message

index 3779f77..2a190e0 100644
@@ -42,6 +42,8 @@ static int rx_ringlen = 256;
 module_param(rx_ringlen, int, 0444);
 static int tx_ringlen = 256;
 module_param(tx_ringlen, int, 0444);
+static int sg_enabled = 1;
+module_param(sg_enabled, int, 0444);
 
 #undef PDEBUG             /* undef it, just in case */
 #ifdef VBUS_ENET_DEBUG

@@ -63,8 +65,17 @@ struct vbus_enet_priv {
 	struct vbus_enet_queue     rxq;
 	struct vbus_enet_queue     txq;
 	struct tasklet_struct      txtask;
+	struct {
+		int                sg:1;
+		int                tso:1;
+		int                ufo:1;
+		int                tso6:1;
+		int                ecn:1;
+	} flags;
 };
 
+static void vbus_enet_tx_reap(struct vbus_enet_priv *priv, int force);
+
 static struct vbus_enet_priv *
 napi_to_priv(struct napi_struct *napi)
 {

@@ -198,6 +209,93 @@ rx_teardown(struct vbus_enet_priv *priv)
 	}
 }
 
+static int
+tx_setup(struct vbus_enet_priv *priv)
+{
+	struct ioq *ioq = priv->txq.queue;
+	struct ioq_iterator iter;
+	int i;
+	int ret;
+
+	if (!priv->flags.sg)
+		/*
+		 * There is nothing to do for a ring that is not using
+		 * scatter-gather
+		 */
+		return 0;
+
+	ret = ioq_iter_init(ioq, &iter, ioq_idxtype_valid, 0);
+	BUG_ON(ret < 0);
+
+	ret = ioq_iter_seek(&iter, ioq_seek_set, 0, 0);
+	BUG_ON(ret < 0);
+
+	/*
+	 * Now populate each descriptor with an empty SG descriptor
+	 */
+	for (i = 0; i < tx_ringlen; i++) {
+		struct venet_sg *vsg;
+		size_t iovlen = sizeof(struct venet_iov) * (MAX_SKB_FRAGS-1);
+		size_t len = sizeof(*vsg) + iovlen;
+
+		vsg = kzalloc(len, GFP_KERNEL);
+		if (!vsg)
+			return -ENOMEM;
+
+		iter.desc->cookie = (u64)vsg;
+		iter.desc->len    = len;
+		iter.desc->ptr    = (u64)__pa(vsg);
+
+		ret = ioq_iter_seek(&iter, ioq_seek_next, 0, 0);
+		BUG_ON(ret < 0);
+	}
+
+	return 0;
+}
+
+static void
+tx_teardown(struct vbus_enet_priv *priv)
+{
+	struct ioq *ioq = priv->txq.queue;
+	struct ioq_iterator iter;
+	int ret;
+
+	/* forcefully free all outstanding transmissions */
+	vbus_enet_tx_reap(priv, 1);
+
+	if (!priv->flags.sg)
+		/*
+		 * There is nothing else to do for a ring that is not using
+		 * scatter-gather
+		 */
+		return;
+
+	ret = ioq_iter_init(ioq, &iter, ioq_idxtype_valid, 0);
+	BUG_ON(ret < 0);
+
+	/* seek to position 0 */
+	ret = ioq_iter_seek(&iter, ioq_seek_set, 0, 0);
+	BUG_ON(ret < 0);
+
+	/*
+	 * free each valid descriptor
+	 */
+	while (iter.desc->cookie) {
+		struct venet_sg *vsg = (struct venet_sg *)iter.desc->cookie;
+
+		iter.desc->valid = 0;
+		wmb();
+
+		iter.desc->ptr = 0;
+		iter.desc->cookie = 0;
+
+		ret = ioq_iter_seek(&iter, ioq_seek_next, 0, 0);
+		BUG_ON(ret < 0);
+
+		kfree(vsg);
+	}
+}
+
 /*
  * Open and close
  */

@@ -402,14 +500,67 @@ vbus_enet_tx_start(struct sk_buff *skb, struct net_device *dev)
 	BUG_ON(ret < 0);
 	BUG_ON(iter.desc->sown);
 
-	/*
-	 * We simply put the skb right onto the ring.  We will get an interrupt
-	 * later when the data has been consumed and we can reap the pointers
-	 * at that time
-	 */
-	iter.desc->cookie = (u64)skb;
-	iter.desc->len = (u64)skb->len;
-	iter.desc->ptr = (u64)__pa(skb->data);
+	if (priv->flags.sg) {
+		struct venet_sg *vsg = (struct venet_sg *)iter.desc->cookie;
+		struct scatterlist sgl[MAX_SKB_FRAGS+1];
+		struct scatterlist *sg;
+		int count, maxcount = ARRAY_SIZE(sgl);
+
+		sg_init_table(sgl, maxcount);
+
+		memset(vsg, 0, sizeof(*vsg));
+
+		vsg->cookie = (u64)skb;
+		vsg->len    = skb->len;
+
+		if (skb->ip_summed == CHECKSUM_PARTIAL) {
+			vsg->flags      |= VENET_SG_FLAG_NEEDS_CSUM;
+			vsg->csum.start  = skb->csum_start - skb_headroom(skb);
+			vsg->csum.offset = skb->csum_offset;
+		}
+
+		if (skb_is_gso(skb)) {
+			struct skb_shared_info *sinfo = skb_shinfo(skb);
+
+			vsg->flags |= VENET_SG_FLAG_GSO;
+
+			vsg->gso.hdrlen = skb_transport_header(skb) - skb->data;
+			vsg->gso.size = sinfo->gso_size;
+			if (sinfo->gso_type & SKB_GSO_TCPV4)
+				vsg->gso.type = VENET_GSO_TYPE_TCPV4;
+			else if (sinfo->gso_type & SKB_GSO_TCPV6)
+				vsg->gso.type = VENET_GSO_TYPE_TCPV6;
+			else if (sinfo->gso_type & SKB_GSO_UDP)
+				vsg->gso.type = VENET_GSO_TYPE_UDP;
+			else
+				panic("Virtual-Ethernet: unknown GSO type " \
+				      "0x%x\n", sinfo->gso_type);
+
+			if (sinfo->gso_type & SKB_GSO_TCP_ECN)
+				vsg->flags |= VENET_SG_FLAG_ECN;
+		}
+
+		count = skb_to_sgvec(skb, sgl, 0, skb->len);
+
+		BUG_ON(count > maxcount);
+
+		for (sg = &sgl[0]; sg; sg = sg_next(sg)) {
+			struct venet_iov *iov = &vsg->iov[vsg->count++];
+
+			iov->len = sg->length;
+			iov->ptr = (u64)sg_phys(sg);
+		}
+
+	} else {
+		/*
+		 * non scatter-gather mode: simply put the skb right onto the
+		 * ring.
+		 */
+		iter.desc->cookie = (u64)skb;
+		iter.desc->len = (u64)skb->len;
+		iter.desc->ptr = (u64)__pa(skb->data);
+	}
+
 	iter.desc->valid  = 1;
 
 	priv->dev->stats.tx_packets++;

@@ -465,7 +616,17 @@ vbus_enet_tx_reap(struct vbus_enet_priv *priv, int force)
 	 * owned by the south-side
 	 */
 	while (iter.desc->valid && (!iter.desc->sown || force)) {
-		struct sk_buff *skb = (struct sk_buff *)iter.desc->cookie;
+		struct sk_buff *skb;
+
+		if (priv->flags.sg) {
+			struct venet_sg *vsg;
+
+			vsg = (struct venet_sg *)iter.desc->cookie;
+			skb = (struct sk_buff *)vsg->cookie;
+
+		} else {
+			skb = (struct sk_buff *)iter.desc->cookie;
+		}
 
 		PDEBUG("%lld: completed sending %d bytes\n",
 		       priv->vdev->id, skb->len);

@@ -546,6 +707,47 @@ tx_isr(struct ioq_notifier *notifier)
        tasklet_schedule(&priv->txtask);
 }
 
+static int
+vbus_enet_negcap(struct vbus_enet_priv *priv)
+{
+	int ret;
+	struct venet_capabilities caps;
+
+	memset(&caps, 0, sizeof(caps));
+
+	if (sg_enabled) {
+		caps.gid = VENET_CAP_GROUP_SG;
+		caps.bits |= (VENET_CAP_SG|VENET_CAP_TSO4|VENET_CAP_TSO6
+			      |VENET_CAP_ECN|VENET_CAP_UFO);
+	}
+
+	ret = devcall(priv, VENET_FUNC_NEGCAP, &caps, sizeof(caps));
+	if (ret < 0)
+		return ret;
+
+	if (caps.bits & VENET_CAP_SG) {
+		priv->flags.sg = true;
+
+		if (caps.bits & VENET_CAP_TSO4)
+			priv->flags.tso = true;
+		if (caps.bits & VENET_CAP_TSO6)
+			priv->flags.tso6 = true;
+		if (caps.bits & VENET_CAP_UFO)
+			priv->flags.ufo = true;
+		if (caps.bits & VENET_CAP_ECN)
+			priv->flags.ecn = true;
+
+		printk(KERN_INFO "VBUSENET %lld: " \
+		       "Detected GSO features %s%s%s%s\n", priv->vdev->id,
+		       priv->flags.tso  ? "t" : "-",
+		       priv->flags.tso6 ? "T" : "-",
+		       priv->flags.ufo  ? "u" : "-",
+		       priv->flags.ecn  ? "e" : "-");
+	}
+
+	return 0;
+}
+
 static const struct net_device_ops vbus_enet_netdev_ops = {
 	.ndo_open          = vbus_enet_open,
 	.ndo_stop          = vbus_enet_stop,

@@ -582,12 +784,21 @@ vbus_enet_probe(struct vbus_device_proxy *vdev)
 	priv->dev  = dev;
 	priv->vdev = vdev;
 
+	ret = vbus_enet_negcap(priv);
+	if (ret < 0) {
+		printk(KERN_INFO "VENET: Error negotiating capabilities for " \
+		       "%lld\n",
+		       priv->vdev->id);
+		goto out_free;
+	}
+
 	tasklet_init(&priv->txtask, deferred_tx_isr, (unsigned long)priv);
 
 	queue_init(priv, &priv->rxq, VENET_QUEUE_RX, rx_ringlen, rx_isr);
 	queue_init(priv, &priv->txq, VENET_QUEUE_TX, tx_ringlen, tx_isr);
 
 	rx_setup(priv);
+	tx_setup(priv);
 
 	ioq_notify_enable(priv->rxq.queue, 0);  /* enable interrupts */
 	ioq_notify_enable(priv->txq.queue, 0);

@@ -607,6 +818,22 @@ vbus_enet_probe(struct vbus_device_proxy *vdev)
 
 	dev->features |= NETIF_F_HIGHDMA;
 
+	if (priv->flags.sg) {
+		dev->features |= NETIF_F_SG|NETIF_F_HW_CSUM|NETIF_F_FRAGLIST;
+
+		if (priv->flags.tso)
+			dev->features |= NETIF_F_TSO;
+
+		if (priv->flags.ufo)
+			dev->features |= NETIF_F_UFO;
+
+		if (priv->flags.tso6)
+			dev->features |= NETIF_F_TSO6;
+
+		if (priv->flags.ecn)
+			dev->features |= NETIF_F_TSO_ECN;
+	}
+
 	ret = register_netdev(dev);
 	if (ret < 0) {
 		printk(KERN_INFO "VENET: error %i registering device \"%s\"\n",

@@ -634,9 +861,9 @@ vbus_enet_remove(struct vbus_device_proxy *vdev)
 	napi_disable(&priv->napi);
 
 	rx_teardown(priv);
-	vbus_enet_tx_reap(priv, 1);
-
 	ioq_put(priv->rxq.queue);
+
+	tx_teardown(priv);
 	ioq_put(priv->txq.queue);
 
 	dev->ops->close(dev, 0);

index ef6b199..1c96b90 100644
@@ -35,8 +35,43 @@ struct venet_capabilities {
 	__u32 bits;
 };
 
-/* CAPABILITIES-GROUP 0 */
-/* #define VENET_CAP_FOO    0   (No capabilities defined yet, for now) */
+#define VENET_CAP_GROUP_SG 0
+
+/* CAPABILITIES-GROUP SG */
+#define VENET_CAP_SG     (1 << 0)
+#define VENET_CAP_TSO4   (1 << 1)
+#define VENET_CAP_TSO6   (1 << 2)
+#define VENET_CAP_ECN    (1 << 3)
+#define VENET_CAP_UFO    (1 << 4)
+
+struct venet_iov {
+	__u32 len;
+	__u64 ptr;
+};
+
+#define VENET_SG_FLAG_NEEDS_CSUM (1 << 0)
+#define VENET_SG_FLAG_GSO        (1 << 1)
+#define VENET_SG_FLAG_ECN        (1 << 2)
+
+struct venet_sg {
+	__u64            cookie;
+	__u32            flags;
+	__u32            len;     /* total length of all iovs */
+	struct {
+		__u16    start;	  /* csum starting position */
+		__u16    offset;  /* offset to place csum */
+	} csum;
+	struct {
+#define VENET_GSO_TYPE_TCPV4	0	/* IPv4 TCP (TSO) */
+#define VENET_GSO_TYPE_UDP	1	/* IPv4 UDP (UFO) */
+#define VENET_GSO_TYPE_TCPV6	2	/* IPv6 TCP */
+		__u8     type;
+		__u16    hdrlen;
+		__u16    size;
+	} gso;
+	__u32            count;   /* nr of iovs */
+	struct venet_iov iov[1];
+};
 
 #define VENET_FUNC_LINKUP   0
 #define VENET_FUNC_LINKDOWN 1

index 148e2c8..a1f2dc6 100644
@@ -81,6 +81,13 @@ enum {
 	TX_IOQ_CONGESTED,
 };
 
+struct venettap;
+
+struct venettap_rx_ops {
+	int (*decode)(struct venettap *priv, void *ptr, int len);
+	int (*import)(struct venettap *, struct sk_buff *, void *, int);
+};
+
 struct venettap {
 	spinlock_t                   lock;
 	unsigned char                hmac[ETH_ALEN]; /* host-mac */

@@ -108,7 +115,13 @@ struct venettap {
 		struct vbus_memctx          *ctx;
 		struct venettap_queue        rxq;
 		struct venettap_queue        txq;
+		struct venettap_rx_ops      *rx_ops;
 		wait_queue_head_t            rx_empty;
+		struct {
+			struct venet_sg     *desc;
+			size_t               len;
+			int                  enabled:1;
+		} sg;
 		int                          connected:1;
 		int                          opened:1;
 		int                          link:1;

@@ -290,6 +303,183 @@ venettap_change_mtu(struct net_device *dev, int new_mtu)
 }
 
 /*
+ * ---------------------------
+ * Scatter-Gather support
+ * ---------------------------
+ */
+
+/* assumes reference to priv->vbus.conn held */
+static int
+venettap_sg_decode(struct venettap *priv, void *ptr, int len)
+{
+	struct venet_sg *vsg;
+	struct vbus_memctx *ctx;
+	int ret;
+
+	/*
+	 * SG is enabled, so we need to pull in the venet_sg
+	 * header before we can interpret the rest of the
+	 * packet
+	 *
+	 * FIXME: Make sure this is not too big
+	 */
+	if (unlikely(len > priv->vbus.sg.len)) {
+		kfree(priv->vbus.sg.desc);
+		priv->vbus.sg.desc = kzalloc(len, GFP_KERNEL);
+	}
+
+	vsg = priv->vbus.sg.desc;
+	ctx = priv->vbus.ctx;
+
+	ret = ctx->ops->copy_from(ctx, vsg, ptr, len);
+	BUG_ON(ret);
+
+	/*
+	 * Non GSO type packets should be constrained by the MTU setting
+	 * on the host
+	 */
+	if (!(vsg->flags & VENET_SG_FLAG_GSO)
+	    && (vsg->len > (priv->netif.dev->mtu + ETH_HLEN)))
+		return -1;
+
+	return vsg->len;
+}
+
+/*
+ * venettap_sg_import - import an skb in scatter-gather mode
+ *
+ * assumes reference to priv->vbus.conn held
+ */
+static int
+venettap_sg_import(struct venettap *priv, struct sk_buff *skb,
+		   void *ptr, int len)
+{
+	struct venet_sg *vsg = priv->vbus.sg.desc;
+	struct vbus_memctx *ctx = priv->vbus.ctx;
+	int remain = len;
+	int ret;
+	int i;
+
+	PDEBUG("Importing %d bytes in %d segments\n", len, vsg->count);
+
+	for (i = 0; i < vsg->count; i++) {
+		struct venet_iov *iov = &vsg->iov[i];
+
+		if (remain < iov->len)
+			return -EINVAL;
+
+		PDEBUG("Segment %d: %p/%d\n", i, iov->ptr, iov->len);
+
+		ret = ctx->ops->copy_from(ctx, skb_tail_pointer(skb),
+					 (void *)iov->ptr,
+					 iov->len);
+		if (ret)
+			return -EFAULT;
+
+		skb_put(skb, iov->len);
+		remain -= iov->len;
+	}
+
+	if (vsg->flags & VENET_SG_FLAG_NEEDS_CSUM
+	    && !skb_partial_csum_set(skb, vsg->csum.start, vsg->csum.offset))
+		return -EINVAL;
+
+	if (vsg->flags & VENET_SG_FLAG_GSO) {
+		struct skb_shared_info *sinfo = skb_shinfo(skb);
+
+		PDEBUG("GSO packet detected\n");
+
+		switch (vsg->gso.type) {
+		case VENET_GSO_TYPE_TCPV4:
+			sinfo->gso_type = SKB_GSO_TCPV4;
+			break;
+		case VENET_GSO_TYPE_TCPV6:
+			sinfo->gso_type = SKB_GSO_TCPV6;
+			break;
+		case VENET_GSO_TYPE_UDP:
+			sinfo->gso_type = SKB_GSO_UDP;
+			break;
+		default:
+			PDEBUG("Illegal GSO type: %d\n", vsg->gso.type);
+			priv->netif.stats.rx_frame_errors++;
+			kfree_skb(skb);
+			return -EINVAL;
+		}
+
+		if (vsg->flags & VENET_SG_FLAG_ECN)
+			sinfo->gso_type |= SKB_GSO_TCP_ECN;
+
+		sinfo->gso_size = vsg->gso.size;
+		if (skb_shinfo(skb)->gso_size == 0) {
+			PDEBUG("Illegal GSO size: %d\n", vsg->gso.size);
+			priv->netif.stats.rx_frame_errors++;
+			kfree_skb(skb);
+			return -EINVAL;
+		}
+
+		/* Header must be checked, and gso_segs computed. */
+		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
+		skb_shinfo(skb)->gso_segs = 0;
+	}
+
+	return 0;
+}
+
+static struct venettap_rx_ops venettap_sg_rx_ops = {
+	.decode = venettap_sg_decode,
+	.import = venettap_sg_import,
+};
+
+/*
+ * ---------------------------
+ * Flat (non Scatter-Gather) support
+ * ---------------------------
+ */
+
+/* assumes reference to priv->vbus.conn held */
+static int
+venettap_flat_decode(struct venettap *priv, void *ptr, int len)
+{
+	size_t maxlen = priv->netif.dev->mtu + ETH_HLEN;
+
+	if (len > maxlen)
+		return -1;
+
+	/*
+	 * If SG is *not* enabled, the length is simply the
+	 * descriptor length
+	 */
+
+	return len;
+}
+
+/*
+ * venettap_rx_flat - import an skb in non scatter-gather mode
+ *
+ * assumes reference to priv->vbus.conn held
+ */
+static int
+venettap_flat_import(struct venettap *priv, struct sk_buff *skb,
+		     void *ptr, int len)
+{
+	struct vbus_memctx *ctx = priv->vbus.ctx;
+	int ret;
+
+	ret = ctx->ops->copy_from(ctx, skb_tail_pointer(skb), ptr, len);
+	if (ret)
+		return -EFAULT;
+
+	skb_put(skb, len);
+
+	return 0;
+}
+
+static struct venettap_rx_ops venettap_flat_rx_ops = {
+	.decode = venettap_flat_decode,
+	.import = venettap_flat_import,
+};
+
+/*
  * The poll implementation.
  */
 static int

@@ -303,6 +493,7 @@ venettap_rx(struct venettap *priv)
 	int                         ret;
 	unsigned long               flags;
 	struct vbus_connection     *conn;
+	struct venettap_rx_ops     *rx_ops;
 
 	PDEBUG("polling...\n");
 

@@ -326,6 +517,8 @@ venettap_rx(struct venettap *priv)
 	ioq = priv->vbus.rxq.queue;
 	ctx = priv->vbus.ctx;
 
+	rx_ops = priv->vbus.rx_ops;
+
 	spin_unlock_irqrestore(&priv->lock, flags);
 
 	/* We want to iterate on the head of the in-use index */

@@ -340,11 +533,14 @@ venettap_rx(struct venettap *priv)
 	 * the north side
 	 */
 	while (iter.desc->sown) {
-		size_t len = iter.desc->len;
-		size_t maxlen = priv->netif.dev->mtu + ETH_HLEN;
 		struct sk_buff *skb = NULL;
+		size_t len;
+
+		len = rx_ops->decode(priv,
+				     (void *)iter.desc->ptr,
+				     iter.desc->len);
 
-		if (unlikely(len > maxlen)) {
+		if (unlikely(len < 0)) {
 			priv->netif.stats.rx_errors++;
 			priv->netif.stats.rx_length_errors++;
 			goto next;

@@ -362,10 +558,8 @@ venettap_rx(struct venettap *priv)
 		/* align IP on 16B boundary */
 		skb_reserve(skb, 2);
 
-		ret = ctx->ops->copy_from(ctx, skb->data,
-					 (void *)iter.desc->ptr,
-					 len);
-		if (unlikely(ret)) {
+		ret = rx_ops->import(priv, skb, (void *)iter.desc->ptr, len);
+		if (unlikely(ret < 0)) {
 			priv->netif.stats.rx_errors++;
 			goto next;
 		}

@@ -846,6 +1040,23 @@ venettap_macquery(struct venettap *priv, void *data, unsigned long len)
 	return 0;
 }
 
+static u32
+venettap_negcap_sg(struct venettap *priv, u32 requested)
+{
+	u32 available = VENET_CAP_SG|VENET_CAP_TSO4|VENET_CAP_TSO6
+		|VENET_CAP_ECN;
+	u32 ret;
+
+	ret = available & requested;
+
+	if (ret & VENET_CAP_SG) {
+		priv->vbus.sg.enabled = true;
+		priv->vbus.rx_ops = &venettap_sg_rx_ops;
+	}
+
+	return ret;
+}
+
 /*
  * Negotiate Capabilities - This function is provided so that the
  * interface may be extended without breaking ABI compatability

@@ -873,6 +1084,9 @@ venettap_negcap(struct venettap *priv, void *data, unsigned long len)
 		return -EFAULT;
 
 	switch (caps.gid) {
+	case VENET_CAP_GROUP_SG:
+		caps.bits = venettap_negcap_sg(priv, caps.bits);
+		break;
 	default:
 		caps.bits = 0;
 		break;

@@ -1057,6 +1271,12 @@ venettap_vlink_release(struct vbus_connection *conn)
 	vbus_memctx_put(priv->vbus.ctx);
 
 	kobject_put(priv->vbus.dev.kobj);
+
+	priv->vbus.sg.enabled = false;
+	priv->vbus.rx_ops = &venettap_flat_rx_ops;
+	kfree(priv->vbus.sg.desc);
+	priv->vbus.sg.desc = NULL;
+	priv->vbus.sg.len = 0;
 }
 
 static struct vbus_connection_ops venettap_vbus_link_ops = {

@@ -1336,6 +1556,7 @@ venettap_device_create(struct vbus_devclass *dc,
 	_vdev->ops             = &venettap_device_ops;
 	_vdev->attrs           = &venettap_attr_group;
 
+	priv->vbus.rx_ops      = &venettap_flat_rx_ops;
 	init_waitqueue_head(&priv->vbus.rx_empty);
 
 	/*

index 592688e..b1726d8 100644
@@ -40,6 +40,12 @@ extern unsigned int do_IRQ(struct pt_regs *regs);
 extern void init_IRQ(void);
 extern void native_init_IRQ(void);
 
+#ifdef CONFIG_SMP
+extern int set_irq_affinity(int irq, cpumask_t mask);
+#endif
+
+extern int irq_to_vector(int irq);
+
 /* Interrupt vector management */
 extern DECLARE_BITMAP(used_vectors, NR_VECTORS);
 extern int vector_used_by_percpu_irq(unsigned int vector);

index bc7ac4d..86a2c36 100644
@@ -614,6 +614,14 @@ set_ioapic_affinity_irq(unsigned int irq, const struct cpumask *mask)
 
 	set_ioapic_affinity_irq_desc(desc, mask);
 }
+
+int set_irq_affinity(int irq, cpumask_t mask)
+{
+	set_ioapic_affinity_irq(irq, &mask);
+
+	return 0;
+}
+
 #endif /* CONFIG_SMP */
 
 /*

@@ -3249,6 +3257,23 @@ void destroy_irq(unsigned int irq)
 	spin_unlock_irqrestore(&vector_lock, flags);
 }
 
+int irq_to_vector(int irq)
+{
+	struct irq_cfg *cfg;
+	unsigned long flags;
+	int ret = -ENOENT;
+
+	spin_lock_irqsave(&vector_lock, flags);
+
+	cfg = irq_cfg(irq);
+	if (cfg && cfg->vector != 0)
+		ret = cfg->vector;
+
+	spin_unlock_irqrestore(&vector_lock, flags);
+
+	return ret;
+}
+
 /*
  * MSI message composition
  */

index 758b7a1..9b0a649 100644
@@ -971,6 +971,7 @@ int kvm_dev_ioctl_check_extension(long ext)
 	case KVM_CAP_NOP_IO_DELAY:
 	case KVM_CAP_MP_STATE:
 	case KVM_CAP_SYNC_MMU:
+	case KVM_CAP_RESET:
 		r = 1;
 		break;
 	case KVM_CAP_COALESCED_MMIO:

index 0424326..7ffd8f5 100644
@@ -396,6 +396,7 @@ struct kvm_trace_rec {
 #ifdef __KVM_HAVE_USER_NMI
 #define KVM_CAP_USER_NMI 22
 #endif
+#define KVM_CAP_RESET 23
 
 /*
  * ioctls for VM fds

@@ -429,6 +430,7 @@ struct kvm_trace_rec {
 				   struct kvm_assigned_pci_dev)
 #define KVM_ASSIGN_IRQ _IOR(KVMIO, 0x70, \
 			    struct kvm_assigned_irq)
+#define KVM_RESET	          _IO(KVMIO,  0x67)
 
 /*
  * ioctls for vcpu fds

index bf6f703..506eca1 100644
@@ -17,6 +17,7 @@
 #include <linux/preempt.h>
 #include <linux/marker.h>
 #include <linux/msi.h>
+#include <linux/notifier.h>
 #include <asm/signal.h>
 
 #include <linux/kvm.h>

@@ -132,6 +133,8 @@ struct kvm {
 	unsigned long mmu_notifier_seq;
 	long mmu_notifier_count;
 #endif
+
+	struct raw_notifier_head reset_notifier; /* triggers when VM reboots */
 };
 
 /* The guest did something we don't support. */

@@ -158,6 +161,9 @@ void kvm_exit(void);
 void kvm_get_kvm(struct kvm *kvm);
 void kvm_put_kvm(struct kvm *kvm);
 
+int kvm_reset_notifier_register(struct kvm *kvm, struct notifier_block *nb);
+int kvm_reset_notifier_unregister(struct kvm *kvm, struct notifier_block *nb);
+
 #define HPA_MSB ((sizeof(hpa_t) * 8) - 1)
 #define HPA_ERR_MASK ((hpa_t)1 << HPA_MSB)
 static inline int is_error_hpa(hpa_t hpa) { return hpa >> HPA_MSB; }

index 29a667c..fca2d25 100644
@@ -868,6 +868,8 @@ static struct kvm *kvm_create_vm(void)
 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
 	kvm_coalesced_mmio_init(kvm);
 #endif
+	RAW_INIT_NOTIFIER_HEAD(&kvm->reset_notifier);
+
 out:
 	return kvm;
 }

@@ -1485,6 +1487,35 @@ void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
 	}
 }
 
+static void kvm_notify_reset(struct kvm *kvm)
+{
+	mutex_lock(&kvm->lock);
+	raw_notifier_call_chain(&kvm->reset_notifier, 0, kvm);
+	mutex_unlock(&kvm->lock);
+}
+
+int kvm_reset_notifier_register(struct kvm *kvm, struct notifier_block *nb)
+{
+	int ret;
+
+	mutex_lock(&kvm->lock);
+	ret = raw_notifier_chain_register(&kvm->reset_notifier, nb);
+	mutex_unlock(&kvm->lock);
+
+	return ret;
+}
+
+int kvm_reset_notifier_unregister(struct kvm *kvm, struct notifier_block *nb)
+{
+	int ret;
+
+	mutex_lock(&kvm->lock);
+	ret = raw_notifier_chain_unregister(&kvm->reset_notifier, nb);
+	mutex_unlock(&kvm->lock);
+
+	return ret;
+}
+
 /*
  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
  */

@@ -1929,6 +1960,11 @@ static long kvm_vm_ioctl(struct file *filp,
 		break;
 	}
 #endif
+	case KVM_RESET: {
+		kvm_notify_reset(kvm);
+		r = 0;
+		break;
+	}
 	default:
 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
 	}

index 730843d..9ae398a 100644
@@ -346,6 +346,12 @@ struct kvm_mem_alias {
 	gfn_t target_gfn;
 };
 
+struct kvm_dynirq {
+	spinlock_t lock;
+	struct rb_root map;
+	struct kvm *kvm;
+};
+
 struct kvm_arch{
 	int naliases;
 	struct kvm_mem_alias aliases[KVM_ALIAS_SLOTS];

@@ -363,6 +369,7 @@ struct kvm_arch{
 	struct iommu_domain *iommu_domain;
 	struct kvm_pic *vpic;
 	struct kvm_ioapic *vioapic;
+	struct kvm_dynirq *dynirq;
 	struct kvm_pit *vpit;
 	struct hlist_head irq_ack_notifier_list;
 	int vapics_in_nmi_mode;

@@ -519,6 +526,8 @@ int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
 			  const void *val, int bytes);
 int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
 		  gpa_t addr, unsigned long *ret);
+int kvm_dynirq_hc(struct kvm_vcpu *vcpu, int nr, gpa_t gpa, size_t len);
+void kvm_free_dynirq(struct kvm *kvm);
 
 extern bool tdp_enabled;
 

index b8a3305..fba210e 100644
@@ -13,6 +13,7 @@
 #define KVM_FEATURE_CLOCKSOURCE		0
 #define KVM_FEATURE_NOP_IO_DELAY	1
 #define KVM_FEATURE_MMU_OP		2
+#define KVM_FEATURE_DYNIRQ		3
 
 #define MSR_KVM_WALL_CLOCK  0x11
 #define MSR_KVM_SYSTEM_TIME 0x12

@@ -45,6 +46,16 @@ struct kvm_mmu_op_release_pt {
 	__u64 pt_phys;
 };
 
+/* Operations for KVM_HC_DYNIRQ */
+#define KVM_DYNIRQ_OP_SET   1
+#define KVM_DYNIRQ_OP_CLEAR 2
+
+struct kvm_dynirq_set {
+	__u32 irq;
+	__u32 vec;  /* x86 IDT vector */
+	__u32 dest; /* 0-based vcpu id */
+};
+
 #ifdef __KERNEL__
 #include <asm/processor.h>
 

329 lines not on this page · raw message