sgi-xp: setup the notify GRU message queue
Dean Nelson [Wed, 30 Jul 2008 05:34:19 +0000 (22:34 -0700)]
Setup the notify GRU message queue that is used for sending user messages
on UV systems.

Signed-off-by: Dean Nelson <dcn@sgi.com>
Cc: Jack Steiner <steiner@sgi.com>
Cc: "Luck, Tony" <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

drivers/misc/sgi-xp/xp.h
drivers/misc/sgi-xp/xp_main.c
drivers/misc/sgi-xp/xpc.h
drivers/misc/sgi-xp/xpc_channel.c
drivers/misc/sgi-xp/xpc_main.c
drivers/misc/sgi-xp/xpc_sn2.c
drivers/misc/sgi-xp/xpc_uv.c
drivers/misc/sgi-xp/xpnet.c

index 9ac5758..859a528 100644 (file)
 #endif
 
 /*
- * The format of an XPC message is as follows:
- *
- *      +-------+--------------------------------+
- *      | flags |////////////////////////////////|
- *      +-------+--------------------------------+
- *      |             message #                  |
- *      +----------------------------------------+
- *      |     payload (user-defined message)     |
- *      |                                        |
- *                     :
- *      |                                        |
- *      +----------------------------------------+
- *
- * The size of the payload is defined by the user via xpc_connect(). A user-
- * defined message resides in the payload area.
- *
- * The size of a message entry (within a message queue) must be a cacheline
- * sized multiple in order to facilitate the BTE transfer of messages from one
- * message queue to another. A macro, XPC_MSG_SIZE(), is provided for the user
+ * Define macro, XPC_MSG_SIZE(), is provided for the user
  * that wants to fit as many msg entries as possible in a given memory size
  * (e.g. a memory page).
  */
-struct xpc_msg {
-       u8 flags;               /* FOR XPC INTERNAL USE ONLY */
-       u8 reserved[7];         /* FOR XPC INTERNAL USE ONLY */
-       s64 number;             /* FOR XPC INTERNAL USE ONLY */
-
-       u64 payload;            /* user defined portion of message */
-};
+#define XPC_MSG_MAX_SIZE       128
+#define XPC_MSG_HDR_MAX_SIZE   16
+#define XPC_MSG_PAYLOAD_MAX_SIZE (XPC_MSG_MAX_SIZE - XPC_MSG_HDR_MAX_SIZE)
 
-#define XPC_MSG_PAYLOAD_OFFSET (u64) (&((struct xpc_msg *)0)->payload)
 #define XPC_MSG_SIZE(_payload_size) \
-               L1_CACHE_ALIGN(XPC_MSG_PAYLOAD_OFFSET + (_payload_size))
+                               ALIGN(XPC_MSG_HDR_MAX_SIZE + (_payload_size), \
+                                     is_uv() ? 64 : 128)
+
 
 /*
  * Define the return values and values passed to user's callout functions.
@@ -210,7 +189,10 @@ enum xp_retval {
        xpGruCopyError,         /* 58: gru_copy_gru() returned error */
        xpGruSendMqError,       /* 59: gru send message queue related error */
 
-       xpUnknownReason         /* 60: unknown reason - must be last in enum */
+       xpBadChannelNumber,     /* 60: invalid channel number */
+       xpBadMsgType,           /* 60: invalid message type */
+
+       xpUnknownReason         /* 61: unknown reason - must be last in enum */
 };
 
 /*
@@ -261,6 +243,9 @@ typedef void (*xpc_channel_func) (enum xp_retval reason, short partid,
  * calling xpc_received().
  *
  * All other reason codes indicate failure.
+ *
+ * NOTE: The user defined function must be callable by an interrupt handler
+ *       and thus cannot block.
  */
 typedef void (*xpc_notify_func) (enum xp_retval reason, short partid,
                                 int ch_number, void *key);
@@ -284,7 +269,7 @@ struct xpc_registration {
        xpc_channel_func func;  /* function to call */
        void *key;              /* pointer to user's key */
        u16 nentries;           /* #of msg entries in local msg queue */
-       u16 msg_size;           /* message queue's message size */
+       u16 entry_size;         /* message queue's message entry size */
        u32 assigned_limit;     /* limit on #of assigned kthreads */
        u32 idle_limit;         /* limit on #of idle kthreads */
 } ____cacheline_aligned;
index f86ad3a..66a1d19 100644 (file)
@@ -154,6 +154,9 @@ xpc_connect(int ch_number, xpc_channel_func func, void *key, u16 payload_size,
        DBUG_ON(func == NULL);
        DBUG_ON(assigned_limit == 0 || idle_limit > assigned_limit);
 
+       if (XPC_MSG_SIZE(payload_size) > XPC_MSG_MAX_SIZE)
+               return xpPayloadTooBig;
+
        registration = &xpc_registrations[ch_number];
 
        if (mutex_lock_interruptible(&registration->mutex) != 0)
@@ -166,7 +169,7 @@ xpc_connect(int ch_number, xpc_channel_func func, void *key, u16 payload_size,
        }
 
        /* register the channel for connection */
-       registration->msg_size = XPC_MSG_SIZE(payload_size);
+       registration->entry_size = XPC_MSG_SIZE(payload_size);
        registration->nentries = nentries;
        registration->assigned_limit = assigned_limit;
        registration->idle_limit = idle_limit;
@@ -220,7 +223,7 @@ xpc_disconnect(int ch_number)
        registration->func = NULL;
        registration->key = NULL;
        registration->nentries = 0;
-       registration->msg_size = 0;
+       registration->entry_size = 0;
        registration->assigned_limit = 0;
        registration->idle_limit = 0;
 
index 4c26181..619208d 100644 (file)
@@ -181,8 +181,8 @@ struct xpc_vars_part_sn2 {
                                  xpc_nasid_mask_nlongs))
 
 /*
- * The activate_mq is used to send/receive messages that affect XPC's heartbeat,
- * partition active state, and channel state. This is UV only.
+ * The activate_mq is used to send/receive GRU messages that affect XPC's
+ * heartbeat, partition active state, and channel state. This is UV only.
  */
 struct xpc_activate_mq_msghdr_uv {
        short partid;           /* sender's partid */
@@ -209,45 +209,45 @@ struct xpc_activate_mq_msghdr_uv {
 #define XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV         11
 
 struct xpc_activate_mq_msg_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
 };
 
 struct xpc_activate_mq_msg_heartbeat_req_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        u64 heartbeat;
 };
 
 struct xpc_activate_mq_msg_activate_req_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        unsigned long rp_gpa;
        unsigned long activate_mq_gpa;
 };
 
 struct xpc_activate_mq_msg_deactivate_req_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        enum xp_retval reason;
 };
 
 struct xpc_activate_mq_msg_chctl_closerequest_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        short ch_number;
        enum xp_retval reason;
 };
 
 struct xpc_activate_mq_msg_chctl_closereply_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        short ch_number;
 };
 
 struct xpc_activate_mq_msg_chctl_openrequest_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        short ch_number;
-       short msg_size;         /* size of notify_mq's messages */
+       short entry_size;       /* size of notify_mq's GRU messages */
        short local_nentries;   /* ??? Is this needed? What is? */
 };
 
 struct xpc_activate_mq_msg_chctl_openreply_uv {
-       struct xpc_activate_mq_msghdr_uv header;
+       struct xpc_activate_mq_msghdr_uv hdr;
        short ch_number;
        short remote_nentries;  /* ??? Is this needed? What is? */
        short local_nentries;   /* ??? Is this needed? What is? */
@@ -284,7 +284,7 @@ struct xpc_gp_sn2 {
  */
 struct xpc_openclose_args {
        u16 reason;             /* reason why channel is closing */
-       u16 msg_size;           /* sizeof each message entry */
+       u16 entry_size;         /* sizeof each message entry */
        u16 remote_nentries;    /* #of message entries in remote msg queue */
        u16 local_nentries;     /* #of message entries in local msg queue */
        unsigned long local_msgqueue_pa; /* phys addr of local message queue */
@@ -294,22 +294,79 @@ struct xpc_openclose_args {
              L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * \
              XPC_MAX_NCHANNELS)
 
-/* struct xpc_msg flags */
 
-#define        XPC_M_DONE              0x01    /* msg has been received/consumed */
-#define        XPC_M_READY             0x02    /* msg is ready to be sent */
-#define        XPC_M_INTERRUPT         0x04    /* send interrupt when msg consumed */
+/*
+ * Structures to define a fifo singly-linked list.
+ */
+
+struct xpc_fifo_entry_uv {
+       struct xpc_fifo_entry_uv *next;
+};
+
+struct xpc_fifo_head_uv {
+       struct xpc_fifo_entry_uv *first;
+       struct xpc_fifo_entry_uv *last;
+       spinlock_t lock;
+       int n_entries;
+};
+
+/*
+ * Define a sn2 styled message.
+ *
+ * A user-defined message resides in the payload area. The max size of the
+ * payload is defined by the user via xpc_connect().
+ *
+ * The size of a message entry (within a message queue) must be a 128-byte
+ * cacheline sized multiple in order to facilitate the BTE transfer of messages
+ * from one message queue to another.
+ */
+struct xpc_msg_sn2 {
+       u8 flags;               /* FOR XPC INTERNAL USE ONLY */
+       u8 reserved[7];         /* FOR XPC INTERNAL USE ONLY */
+       s64 number;             /* FOR XPC INTERNAL USE ONLY */
+
+       u64 payload;            /* user defined portion of message */
+};
+
+/* struct xpc_msg_sn2 flags */
+
+#define        XPC_M_SN2_DONE          0x01    /* msg has been received/consumed */
+#define        XPC_M_SN2_READY         0x02    /* msg is ready to be sent */
+#define        XPC_M_SN2_INTERRUPT     0x04    /* send interrupt when msg consumed */
+
+/*
+ * The format of a uv XPC notify_mq GRU message is as follows:
+ *
+ * A user-defined message resides in the payload area. The max size of the
+ * payload is defined by the user via xpc_connect().
+ *
+ * The size of a message (payload and header) sent via the GRU must be either 1
+ * or 2 GRU_CACHE_LINE_BYTES in length.
+ */
 
-#define XPC_MSG_ADDRESS(_payload) \
-               ((struct xpc_msg *)((u8 *)(_payload) - XPC_MSG_PAYLOAD_OFFSET))
+struct xpc_notify_mq_msghdr_uv {
+       union {
+               unsigned int gru_msg_hdr;       /* FOR GRU INTERNAL USE ONLY */
+               struct xpc_fifo_entry_uv next;  /* FOR XPC INTERNAL USE ONLY */
+       } u;
+       short partid;           /* FOR XPC INTERNAL USE ONLY */
+       u8 ch_number;           /* FOR XPC INTERNAL USE ONLY */
+       u8 size;                /* FOR XPC INTERNAL USE ONLY */
+       unsigned int msg_slot_number;   /* FOR XPC INTERNAL USE ONLY */
+};
+
+struct xpc_notify_mq_msg_uv {
+       struct xpc_notify_mq_msghdr_uv hdr;
+       unsigned long payload;
+};
 
 /*
- * Defines notify entry.
+ * Define sn2's notify entry.
  *
  * This is used to notify a message's sender that their message was received
  * and consumed by the intended recipient.
  */
-struct xpc_notify {
+struct xpc_notify_sn2 {
        u8 type;                /* type of notification */
 
        /* the following two fields are only used if type == XPC_N_CALL */
@@ -317,9 +374,20 @@ struct xpc_notify {
        void *key;              /* pointer to user's key */
 };
 
-/* struct xpc_notify type of notification */
+/* struct xpc_notify_sn2 type of notification */
 
-#define        XPC_N_CALL              0x01    /* notify function provided by user */
+#define        XPC_N_CALL      0x01    /* notify function provided by user */
+
+/*
+ * Define uv's version of the notify entry. It additionally is used to allocate
+ * a msg slot on the remote partition into which is copied a sent message.
+ */
+struct xpc_send_msg_slot_uv {
+       struct xpc_fifo_entry_uv next;
+       unsigned int msg_slot_number;
+       xpc_notify_func func;   /* user's notify function */
+       void *key;              /* pointer to user's key */
+};
 
 /*
  * Define the structure that manages all the stuff required by a channel. In
@@ -409,14 +477,14 @@ struct xpc_channel_sn2 {
                                             /* opening or closing of channel */
 
        void *local_msgqueue_base;      /* base address of kmalloc'd space */
-       struct xpc_msg *local_msgqueue; /* local message queue */
+       struct xpc_msg_sn2 *local_msgqueue;     /* local message queue */
        void *remote_msgqueue_base;     /* base address of kmalloc'd space */
-       struct xpc_msg *remote_msgqueue; /* cached copy of remote partition's */
-                                        /* local message queue */
+       struct xpc_msg_sn2 *remote_msgqueue; /* cached copy of remote */
+                                          /* partition's local message queue */
        unsigned long remote_msgqueue_pa; /* phys addr of remote partition's */
                                          /* local message queue */
 
-       struct xpc_notify *notify_queue;    /* notify queue for messages sent */
+       struct xpc_notify_sn2 *notify_queue;/* notify queue for messages sent */
 
        /* various flavors of local and remote Get/Put values */
 
@@ -432,6 +500,12 @@ struct xpc_channel_sn2 {
 struct xpc_channel_uv {
        unsigned long remote_notify_mq_gpa;     /* gru phys address of remote */
                                                /* partition's notify mq */
+
+       struct xpc_send_msg_slot_uv *send_msg_slots;
+       struct xpc_notify_mq_msg_uv *recv_msg_slots;
+
+       struct xpc_fifo_head_uv msg_slot_free_list;
+       struct xpc_fifo_head_uv recv_msg_list;  /* deliverable payloads */
 };
 
 struct xpc_channel {
@@ -444,7 +518,7 @@ struct xpc_channel {
 
        u16 number;             /* channel # */
 
-       u16 msg_size;           /* sizeof each msg entry */
+       u16 entry_size;         /* sizeof each msg entry */
        u16 local_nentries;     /* #of msg entries in local msg queue */
        u16 remote_nentries;    /* #of msg entries in remote msg queue */
 
@@ -733,8 +807,8 @@ extern enum xp_retval (*xpc_setup_msg_structures) (struct xpc_channel *);
 extern void (*xpc_teardown_msg_structures) (struct xpc_channel *);
 extern void (*xpc_notify_senders_of_disconnect) (struct xpc_channel *);
 extern void (*xpc_process_msg_chctl_flags) (struct xpc_partition *, int);
-extern int (*xpc_n_of_deliverable_msgs) (struct xpc_channel *);
-extern struct xpc_msg *(*xpc_get_deliverable_msg) (struct xpc_channel *);
+extern int (*xpc_n_of_deliverable_payloads) (struct xpc_channel *);
+extern void *(*xpc_get_deliverable_payload) (struct xpc_channel *);
 extern void (*xpc_request_partition_activation) (struct xpc_rsvd_page *,
                                                 unsigned long, int);
 extern void (*xpc_request_partition_reactivation) (struct xpc_partition *);
@@ -762,9 +836,9 @@ extern void (*xpc_send_chctl_openreply) (struct xpc_channel *, unsigned long *);
 extern void (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *,
                                            unsigned long);
 
-extern enum xp_retval (*xpc_send_msg) (struct xpc_channel *, u32, void *, u16,
-                                      u8, xpc_notify_func, void *);
-extern void (*xpc_received_msg) (struct xpc_channel *, struct xpc_msg *);
+extern enum xp_retval (*xpc_send_payload) (struct xpc_channel *, u32, void *,
+                                          u16, u8, xpc_notify_func, void *);
+extern void (*xpc_received_payload) (struct xpc_channel *, void *);
 
 /* found in xpc_sn2.c */
 extern int xpc_init_sn2(void);
@@ -805,7 +879,7 @@ extern enum xp_retval xpc_initiate_send_notify(short, int, u32, void *, u16,
 extern void xpc_initiate_received(short, int, void *);
 extern void xpc_process_sent_chctl_flags(struct xpc_partition *);
 extern void xpc_connected_callout(struct xpc_channel *);
-extern void xpc_deliver_msg(struct xpc_channel *);
+extern void xpc_deliver_payload(struct xpc_channel *);
 extern void xpc_disconnect_channel(const int, struct xpc_channel *,
                                   enum xp_retval, unsigned long *);
 extern void xpc_disconnect_callout(struct xpc_channel *, enum xp_retval);
index 73df9fb..9cd2ebe 100644 (file)
@@ -139,7 +139,7 @@ xpc_process_disconnect(struct xpc_channel *ch, unsigned long *irq_flags)
 
        ch->func = NULL;
        ch->key = NULL;
-       ch->msg_size = 0;
+       ch->entry_size = 0;
        ch->local_nentries = 0;
        ch->remote_nentries = 0;
        ch->kthreads_assigned_limit = 0;
@@ -315,9 +315,9 @@ again:
 
        if (chctl_flags & XPC_CHCTL_OPENREQUEST) {
 
-               dev_dbg(xpc_chan, "XPC_CHCTL_OPENREQUEST (msg_size=%d, "
+               dev_dbg(xpc_chan, "XPC_CHCTL_OPENREQUEST (entry_size=%d, "
                        "local_nentries=%d) received from partid=%d, "
-                       "channel=%d\n", args->msg_size, args->local_nentries,
+                       "channel=%d\n", args->entry_size, args->local_nentries,
                        ch->partid, ch->number);
 
                if (part->act_state == XPC_P_AS_DEACTIVATING ||
@@ -338,10 +338,10 @@ again:
 
                /*
                 * The meaningful OPENREQUEST connection state fields are:
-                *      msg_size = size of channel's messages in bytes
+                *      entry_size = size of channel's messages in bytes
                 *      local_nentries = remote partition's local_nentries
                 */
-               if (args->msg_size == 0 || args->local_nentries == 0) {
+               if (args->entry_size == 0 || args->local_nentries == 0) {
                        /* assume OPENREQUEST was delayed by mistake */
                        spin_unlock_irqrestore(&ch->lock, irq_flags);
                        return;
@@ -351,14 +351,14 @@ again:
                ch->remote_nentries = args->local_nentries;
 
                if (ch->flags & XPC_C_OPENREQUEST) {
-                       if (args->msg_size != ch->msg_size) {
+                       if (args->entry_size != ch->entry_size) {
                                XPC_DISCONNECT_CHANNEL(ch, xpUnequalMsgSizes,
                                                       &irq_flags);
                                spin_unlock_irqrestore(&ch->lock, irq_flags);
                                return;
                        }
                } else {
-                       ch->msg_size = args->msg_size;
+                       ch->entry_size = args->entry_size;
 
                        XPC_SET_REASON(ch, 0, 0);
                        ch->flags &= ~XPC_C_DISCONNECTED;
@@ -473,7 +473,7 @@ xpc_connect_channel(struct xpc_channel *ch)
        ch->local_nentries = registration->nentries;
 
        if (ch->flags & XPC_C_ROPENREQUEST) {
-               if (registration->msg_size != ch->msg_size) {
+               if (registration->entry_size != ch->entry_size) {
                        /* the local and remote sides aren't the same */
 
                        /*
@@ -492,7 +492,7 @@ xpc_connect_channel(struct xpc_channel *ch)
                        return xpUnequalMsgSizes;
                }
        } else {
-               ch->msg_size = registration->msg_size;
+               ch->entry_size = registration->entry_size;
 
                XPC_SET_REASON(ch, 0, 0);
                ch->flags &= ~XPC_C_DISCONNECTED;
@@ -859,8 +859,8 @@ xpc_initiate_send(short partid, int ch_number, u32 flags, void *payload,
        DBUG_ON(payload == NULL);
 
        if (xpc_part_ref(part)) {
-               ret = xpc_send_msg(&part->channels[ch_number], flags, payload,
-                                  payload_size, 0, NULL, NULL);
+               ret = xpc_send_payload(&part->channels[ch_number], flags,
+                                      payload, payload_size, 0, NULL, NULL);
                xpc_part_deref(part);
        }
 
@@ -911,23 +911,24 @@ xpc_initiate_send_notify(short partid, int ch_number, u32 flags, void *payload,
        DBUG_ON(func == NULL);
 
        if (xpc_part_ref(part)) {
-               ret = xpc_send_msg(&part->channels[ch_number], flags, payload,
-                                  payload_size, XPC_N_CALL, func, key);
+               ret = xpc_send_payload(&part->channels[ch_number], flags,
+                                      payload, payload_size, XPC_N_CALL, func,
+                                      key);
                xpc_part_deref(part);
        }
        return ret;
 }
 
 /*
- * Deliver a message to its intended recipient.
+ * Deliver a message's payload to its intended recipient.
  */
 void
-xpc_deliver_msg(struct xpc_channel *ch)
+xpc_deliver_payload(struct xpc_channel *ch)
 {
-       struct xpc_msg *msg;
+       void *payload;
 
-       msg = xpc_get_deliverable_msg(ch);
-       if (msg != NULL) {
+       payload = xpc_get_deliverable_payload(ch);
+       if (payload != NULL) {
 
                /*
                 * This ref is taken to protect the payload itself from being
@@ -939,18 +940,16 @@ xpc_deliver_msg(struct xpc_channel *ch)
                atomic_inc(&ch->kthreads_active);
 
                if (ch->func != NULL) {
-                       dev_dbg(xpc_chan, "ch->func() called, msg=0x%p, "
-                               "msg_number=%ld, partid=%d, channel=%d\n",
-                               msg, (signed long)msg->number, ch->partid,
+                       dev_dbg(xpc_chan, "ch->func() called, payload=0x%p "
+                               "partid=%d channel=%d\n", payload, ch->partid,
                                ch->number);
 
                        /* deliver the message to its intended recipient */
-                       ch->func(xpMsgReceived, ch->partid, ch->number,
-                                &msg->payload, ch->key);
+                       ch->func(xpMsgReceived, ch->partid, ch->number, payload,
+                                ch->key);
 
-                       dev_dbg(xpc_chan, "ch->func() returned, msg=0x%p, "
-                               "msg_number=%ld, partid=%d, channel=%d\n",
-                               msg, (signed long)msg->number, ch->partid,
+                       dev_dbg(xpc_chan, "ch->func() returned, payload=0x%p "
+                               "partid=%d channel=%d\n", payload, ch->partid,
                                ch->number);
                }
 
@@ -959,14 +958,11 @@ xpc_deliver_msg(struct xpc_channel *ch)
 }
 
 /*
- * Acknowledge receipt of a delivered message.
- *
- * If a message has XPC_M_INTERRUPT set, send an interrupt to the partition
- * that sent the message.
+ * Acknowledge receipt of a delivered message's payload.
  *
  * This function, although called by users, does not call xpc_part_ref() to
  * ensure that the partition infrastructure is in place. It relies on the
- * fact that we called xpc_msgqueue_ref() in xpc_deliver_msg().
+ * fact that we called xpc_msgqueue_ref() in xpc_deliver_payload().
  *
  * Arguments:
  *
@@ -980,14 +976,13 @@ xpc_initiate_received(short partid, int ch_number, void *payload)
 {
        struct xpc_partition *part = &xpc_partitions[partid];
        struct xpc_channel *ch;
-       struct xpc_msg *msg = XPC_MSG_ADDRESS(payload);
 
        DBUG_ON(partid < 0 || partid >= xp_max_npartitions);
        DBUG_ON(ch_number < 0 || ch_number >= part->nchannels);
 
        ch = &part->channels[ch_number];
-       xpc_received_msg(ch, msg);
+       xpc_received_payload(ch, payload);
 
-       /* the call to xpc_msgqueue_ref() was done by xpc_deliver_msg()  */
+       /* the call to xpc_msgqueue_ref() was done by xpc_deliver_payload()  */
        xpc_msgqueue_deref(ch);
 }
index 13ec479..46325fc 100644 (file)
@@ -188,8 +188,8 @@ u64 (*xpc_get_chctl_all_flags) (struct xpc_partition *part);
 enum xp_retval (*xpc_setup_msg_structures) (struct xpc_channel *ch);
 void (*xpc_teardown_msg_structures) (struct xpc_channel *ch);
 void (*xpc_process_msg_chctl_flags) (struct xpc_partition *part, int ch_number);
-int (*xpc_n_of_deliverable_msgs) (struct xpc_channel *ch);
-struct xpc_msg *(*xpc_get_deliverable_msg) (struct xpc_channel *ch);
+int (*xpc_n_of_deliverable_payloads) (struct xpc_channel *ch);
+void *(*xpc_get_deliverable_payload) (struct xpc_channel *ch);
 
 void (*xpc_request_partition_activation) (struct xpc_rsvd_page *remote_rp,
                                          unsigned long remote_rp_pa,
@@ -220,10 +220,11 @@ void (*xpc_send_chctl_openreply) (struct xpc_channel *ch,
 void (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *ch,
                                     unsigned long msgqueue_pa);
 
-enum xp_retval (*xpc_send_msg) (struct xpc_channel *ch, u32 flags,
-                               void *payload, u16 payload_size, u8 notify_type,
-                               xpc_notify_func func, void *key);
-void (*xpc_received_msg) (struct xpc_channel *ch, struct xpc_msg *msg);
+enum xp_retval (*xpc_send_payload) (struct xpc_channel *ch, u32 flags,
+                                   void *payload, u16 payload_size,
+                                   u8 notify_type, xpc_notify_func func,
+                                   void *key);
+void (*xpc_received_payload) (struct xpc_channel *ch, void *payload);
 
 /*
  * Timer function to enforce the timelimit on the partition disengage.
@@ -714,9 +715,9 @@ xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch)
        do {
                /* deliver messages to their intended recipients */
 
-               while (xpc_n_of_deliverable_msgs(ch) > 0 &&
+               while (xpc_n_of_deliverable_payloads(ch) > 0 &&
                       !(ch->flags & XPC_C_DISCONNECTING)) {
-                       xpc_deliver_msg(ch);
+                       xpc_deliver_payload(ch);
                }
 
                if (atomic_inc_return(&ch->kthreads_idle) >
@@ -730,7 +731,7 @@ xpc_kthread_waitmsgs(struct xpc_partition *part, struct xpc_channel *ch)
                        "wait_event_interruptible_exclusive()\n");
 
                (void)wait_event_interruptible_exclusive(ch->idle_wq,
-                               (xpc_n_of_deliverable_msgs(ch) > 0 ||
+                               (xpc_n_of_deliverable_payloads(ch) > 0 ||
                                 (ch->flags & XPC_C_DISCONNECTING)));
 
                atomic_dec(&ch->kthreads_idle);
@@ -775,7 +776,7 @@ xpc_kthread_start(void *args)
                         * additional kthreads to help deliver them. We only
                         * need one less than total #of messages to deliver.
                         */
-                       n_needed = xpc_n_of_deliverable_msgs(ch) - 1;
+                       n_needed = xpc_n_of_deliverable_payloads(ch) - 1;
                        if (n_needed > 0 && !(ch->flags & XPC_C_DISCONNECTING))
                                xpc_activate_kthreads(ch, n_needed);
 
index 8b4b065..b4882cc 100644 (file)
@@ -408,7 +408,7 @@ xpc_send_chctl_openrequest_sn2(struct xpc_channel *ch, unsigned long *irq_flags)
 {
        struct xpc_openclose_args *args = ch->sn.sn2.local_openclose_args;
 
-       args->msg_size = ch->msg_size;
+       args->entry_size = ch->entry_size;
        args->local_nentries = ch->local_nentries;
        XPC_SEND_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_OPENREQUEST, irq_flags);
 }
@@ -1531,14 +1531,14 @@ xpc_allocate_local_msgqueue_sn2(struct xpc_channel *ch)
 
        for (nentries = ch->local_nentries; nentries > 0; nentries--) {
 
-               nbytes = nentries * ch->msg_size;
+               nbytes = nentries * ch->entry_size;
                ch_sn2->local_msgqueue =
                    xpc_kzalloc_cacheline_aligned(nbytes, GFP_KERNEL,
                                                  &ch_sn2->local_msgqueue_base);
                if (ch_sn2->local_msgqueue == NULL)
                        continue;
 
-               nbytes = nentries * sizeof(struct xpc_notify);
+               nbytes = nentries * sizeof(struct xpc_notify_sn2);
                ch_sn2->notify_queue = kzalloc(nbytes, GFP_KERNEL);
                if (ch_sn2->notify_queue == NULL) {
                        kfree(ch_sn2->local_msgqueue_base);
@@ -1578,7 +1578,7 @@ xpc_allocate_remote_msgqueue_sn2(struct xpc_channel *ch)
 
        for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
 
-               nbytes = nentries * ch->msg_size;
+               nbytes = nentries * ch->entry_size;
                ch_sn2->remote_msgqueue =
                    xpc_kzalloc_cacheline_aligned(nbytes, GFP_KERNEL, &ch_sn2->
                                                  remote_msgqueue_base);
@@ -1632,9 +1632,6 @@ xpc_setup_msg_structures_sn2(struct xpc_channel *ch)
 /*
  * Free up message queues and other stuff that were allocated for the specified
  * channel.
- *
- * Note: ch->reason and ch->reason_line are left set for debugging purposes,
- * they're cleared when XPC_C_DISCONNECTED is cleared.
  */
 static void
 xpc_teardown_msg_structures_sn2(struct xpc_channel *ch)
@@ -1674,7 +1671,7 @@ xpc_teardown_msg_structures_sn2(struct xpc_channel *ch)
 static void
 xpc_notify_senders_sn2(struct xpc_channel *ch, enum xp_retval reason, s64 put)
 {
-       struct xpc_notify *notify;
+       struct xpc_notify_sn2 *notify;
        u8 notify_type;
        s64 get = ch->sn.sn2.w_remote_GP.get - 1;
 
@@ -1699,17 +1696,16 @@ xpc_notify_senders_sn2(struct xpc_channel *ch, enum xp_retval reason, s64 put)
                atomic_dec(&ch->n_to_notify);
 
                if (notify->func != NULL) {
-                       dev_dbg(xpc_chan, "notify->func() called, notify=0x%p, "
-                               "msg_number=%ld, partid=%d, channel=%d\n",
+                       dev_dbg(xpc_chan, "notify->func() called, notify=0x%p "
+                               "msg_number=%ld partid=%d channel=%d\n",
                                (void *)notify, get, ch->partid, ch->number);
 
                        notify->func(reason, ch->partid, ch->number,
                                     notify->key);
 
-                       dev_dbg(xpc_chan, "notify->func() returned, "
-                               "notify=0x%p, msg_number=%ld, partid=%d, "
-                               "channel=%d\n", (void *)notify, get,
-                               ch->partid, ch->number);
+                       dev_dbg(xpc_chan, "notify->func() returned, notify=0x%p"
+                               " msg_number=%ld partid=%d channel=%d\n",
+                               (void *)notify, get, ch->partid, ch->number);
                }
        }
 }
@@ -1727,14 +1723,14 @@ static inline void
 xpc_clear_local_msgqueue_flags_sn2(struct xpc_channel *ch)
 {
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg;
+       struct xpc_msg_sn2 *msg;
        s64 get;
 
        get = ch_sn2->w_remote_GP.get;
        do {
-               msg = (struct xpc_msg *)((u64)ch_sn2->local_msgqueue +
-                                        (get % ch->local_nentries) *
-                                        ch->msg_size);
+               msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->local_msgqueue +
+                                            (get % ch->local_nentries) *
+                                            ch->entry_size);
                msg->flags = 0;
        } while (++get < ch_sn2->remote_GP.get);
 }
@@ -1746,24 +1742,30 @@ static inline void
 xpc_clear_remote_msgqueue_flags_sn2(struct xpc_channel *ch)
 {
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg;
+       struct xpc_msg_sn2 *msg;
        s64 put;
 
        put = ch_sn2->w_remote_GP.put;
        do {
-               msg = (struct xpc_msg *)((u64)ch_sn2->remote_msgqueue +
-                                        (put % ch->remote_nentries) *
-                                        ch->msg_size);
+               msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
+                                            (put % ch->remote_nentries) *
+                                            ch->entry_size);
                msg->flags = 0;
        } while (++put < ch_sn2->remote_GP.put);
 }
 
+static int
+xpc_n_of_deliverable_payloads_sn2(struct xpc_channel *ch)
+{
+       return ch->sn.sn2.w_remote_GP.put - ch->sn.sn2.w_local_GP.get;
+}
+
 static void
 xpc_process_msg_chctl_flags_sn2(struct xpc_partition *part, int ch_number)
 {
        struct xpc_channel *ch = &part->channels[ch_number];
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       int nmsgs_sent;
+       int npayloads_sent;
 
        ch_sn2->remote_GP = part->sn.sn2.remote_GPs[ch_number];
 
@@ -1835,7 +1837,7 @@ xpc_process_msg_chctl_flags_sn2(struct xpc_partition *part, int ch_number)
        if (ch_sn2->w_remote_GP.put != ch_sn2->remote_GP.put) {
                /*
                 * Clear msg->flags in previously received messages, so that
-                * they're ready for xpc_get_deliverable_msg().
+                * they're ready for xpc_get_deliverable_payload_sn2().
                 */
                xpc_clear_remote_msgqueue_flags_sn2(ch);
 
@@ -1845,27 +1847,27 @@ xpc_process_msg_chctl_flags_sn2(struct xpc_partition *part, int ch_number)
                        "channel=%d\n", ch_sn2->w_remote_GP.put, ch->partid,
                        ch->number);
 
-               nmsgs_sent = ch_sn2->w_remote_GP.put - ch_sn2->w_local_GP.get;
-               if (nmsgs_sent > 0) {
+               npayloads_sent = xpc_n_of_deliverable_payloads_sn2(ch);
+               if (npayloads_sent > 0) {
                        dev_dbg(xpc_chan, "msgs waiting to be copied and "
                                "delivered=%d, partid=%d, channel=%d\n",
-                               nmsgs_sent, ch->partid, ch->number);
+                               npayloads_sent, ch->partid, ch->number);
 
                        if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)
-                               xpc_activate_kthreads(ch, nmsgs_sent);
+                               xpc_activate_kthreads(ch, npayloads_sent);
                }
        }
 
        xpc_msgqueue_deref(ch);
 }
 
-static struct xpc_msg *
+static struct xpc_msg_sn2 *
 xpc_pull_remote_msg_sn2(struct xpc_channel *ch, s64 get)
 {
        struct xpc_partition *part = &xpc_partitions[ch->partid];
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
        unsigned long remote_msg_pa;
-       struct xpc_msg *msg;
+       struct xpc_msg_sn2 *msg;
        u32 msg_index;
        u32 nmsgs;
        u64 msg_offset;
@@ -1889,13 +1891,13 @@ xpc_pull_remote_msg_sn2(struct xpc_channel *ch, s64 get)
                        nmsgs = ch->remote_nentries - msg_index;
                }
 
-               msg_offset = msg_index * ch->msg_size;
-               msg = (struct xpc_msg *)((u64)ch_sn2->remote_msgqueue +
+               msg_offset = msg_index * ch->entry_size;
+               msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
                    msg_offset);
                remote_msg_pa = ch_sn2->remote_msgqueue_pa + msg_offset;
 
                ret = xpc_pull_remote_cachelines_sn2(part, msg, remote_msg_pa,
-                                                    nmsgs * ch->msg_size);
+                                                    nmsgs * ch->entry_size);
                if (ret != xpSuccess) {
 
                        dev_dbg(xpc_chan, "failed to pull %d msgs starting with"
@@ -1915,26 +1917,21 @@ xpc_pull_remote_msg_sn2(struct xpc_channel *ch, s64 get)
        mutex_unlock(&ch_sn2->msg_to_pull_mutex);
 
        /* return the message we were looking for */
-       msg_offset = (get % ch->remote_nentries) * ch->msg_size;
-       msg = (struct xpc_msg *)((u64)ch_sn2->remote_msgqueue + msg_offset);
+       msg_offset = (get % ch->remote_nentries) * ch->entry_size;
+       msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue + msg_offset);
 
        return msg;
 }
 
-static int
-xpc_n_of_deliverable_msgs_sn2(struct xpc_channel *ch)
-{
-       return ch->sn.sn2.w_remote_GP.put - ch->sn.sn2.w_local_GP.get;
-}
-
 /*
- * Get a message to be delivered.
+ * Get the next deliverable message's payload.
  */
-static struct xpc_msg *
-xpc_get_deliverable_msg_sn2(struct xpc_channel *ch)
+static void *
+xpc_get_deliverable_payload_sn2(struct xpc_channel *ch)
 {
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg = NULL;
+       struct xpc_msg_sn2 *msg;
+       void *payload = NULL;
        s64 get;
 
        do {
@@ -1965,15 +1962,16 @@ xpc_get_deliverable_msg_sn2(struct xpc_channel *ch)
                        msg = xpc_pull_remote_msg_sn2(ch, get);
 
                        DBUG_ON(msg != NULL && msg->number != get);
-                       DBUG_ON(msg != NULL && (msg->flags & XPC_M_DONE));
-                       DBUG_ON(msg != NULL && !(msg->flags & XPC_M_READY));
+                       DBUG_ON(msg != NULL && (msg->flags & XPC_M_SN2_DONE));
+                       DBUG_ON(msg != NULL && !(msg->flags & XPC_M_SN2_READY));
 
+                       payload = &msg->payload;
                        break;
                }
 
        } while (1);
 
-       return msg;
+       return payload;
 }
 
 /*
@@ -1985,7 +1983,7 @@ static void
 xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
 {
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg;
+       struct xpc_msg_sn2 *msg;
        s64 put = initial_put + 1;
        int send_msgrequest = 0;
 
@@ -1995,11 +1993,12 @@ xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
                        if (put == ch_sn2->w_local_GP.put)
                                break;
 
-                       msg = (struct xpc_msg *)((u64)ch_sn2->local_msgqueue +
-                                                (put % ch->local_nentries) *
-                                                ch->msg_size);
+                       msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->
+                                                    local_msgqueue + (put %
+                                                    ch->local_nentries) *
+                                                    ch->entry_size);
 
-                       if (!(msg->flags & XPC_M_READY))
+                       if (!(msg->flags & XPC_M_SN2_READY))
                                break;
 
                        put++;
@@ -2026,7 +2025,7 @@ xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
 
                /*
                 * We need to ensure that the message referenced by
-                * local_GP->put is not XPC_M_READY or that local_GP->put
+                * local_GP->put is not XPC_M_SN2_READY or that local_GP->put
                 * equals w_local_GP.put, so we'll go have a look.
                 */
                initial_put = put;
@@ -2042,10 +2041,10 @@ xpc_send_msgs_sn2(struct xpc_channel *ch, s64 initial_put)
  */
 static enum xp_retval
 xpc_allocate_msg_sn2(struct xpc_channel *ch, u32 flags,
-                    struct xpc_msg **address_of_msg)
+                    struct xpc_msg_sn2 **address_of_msg)
 {
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg;
+       struct xpc_msg_sn2 *msg;
        enum xp_retval ret;
        s64 put;
 
@@ -2097,8 +2096,9 @@ xpc_allocate_msg_sn2(struct xpc_channel *ch, u32 flags,
        }
 
        /* get the message's address and initialize it */
-       msg = (struct xpc_msg *)((u64)ch_sn2->local_msgqueue +
-                                (put % ch->local_nentries) * ch->msg_size);
+       msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->local_msgqueue +
+                                    (put % ch->local_nentries) *
+                                    ch->entry_size);
 
        DBUG_ON(msg->flags != 0);
        msg->number = put;
@@ -2117,20 +2117,20 @@ xpc_allocate_msg_sn2(struct xpc_channel *ch, u32 flags,
  * partition the message is being sent to.
  */
 static enum xp_retval
-xpc_send_msg_sn2(struct xpc_channel *ch, u32 flags, void *payload,
-                u16 payload_size, u8 notify_type, xpc_notify_func func,
-                void *key)
+xpc_send_payload_sn2(struct xpc_channel *ch, u32 flags, void *payload,
+                    u16 payload_size, u8 notify_type, xpc_notify_func func,
+                    void *key)
 {
        enum xp_retval ret = xpSuccess;
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg = msg;
-       struct xpc_notify *notify = notify;
+       struct xpc_msg_sn2 *msg = msg;
+       struct xpc_notify_sn2 *notify = notify;
        s64 msg_number;
        s64 put;
 
        DBUG_ON(notify_type == XPC_N_CALL && func == NULL);
 
-       if (XPC_MSG_SIZE(payload_size) > ch->msg_size)
+       if (XPC_MSG_SIZE(payload_size) > ch->entry_size)
                return xpPayloadTooBig;
 
        xpc_msgqueue_ref(ch);
@@ -2155,7 +2155,7 @@ xpc_send_msg_sn2(struct xpc_channel *ch, u32 flags, void *payload,
                 * Tell the remote side to send an ACK interrupt when the
                 * message has been delivered.
                 */
-               msg->flags |= XPC_M_INTERRUPT;
+               msg->flags |= XPC_M_SN2_INTERRUPT;
 
                atomic_inc(&ch->n_to_notify);
 
@@ -2185,7 +2185,7 @@ xpc_send_msg_sn2(struct xpc_channel *ch, u32 flags, void *payload,
 
        memcpy(&msg->payload, payload, payload_size);
 
-       msg->flags |= XPC_M_READY;
+       msg->flags |= XPC_M_SN2_READY;
 
        /*
         * The preceding store of msg->flags must occur before the following
@@ -2208,12 +2208,15 @@ out_1:
  * Now we actually acknowledge the messages that have been delivered and ack'd
  * by advancing the cached remote message queue's Get value and if requested
  * send a chctl msgrequest to the message sender's partition.
+ *
+ * If a message has XPC_M_SN2_INTERRUPT set, send an interrupt to the partition
+ * that sent the message.
  */
 static void
 xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
 {
        struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
-       struct xpc_msg *msg;
+       struct xpc_msg_sn2 *msg;
        s64 get = initial_get + 1;
        int send_msgrequest = 0;
 
@@ -2223,11 +2226,12 @@ xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
                        if (get == ch_sn2->w_local_GP.get)
                                break;
 
-                       msg = (struct xpc_msg *)((u64)ch_sn2->remote_msgqueue +
-                                                (get % ch->remote_nentries) *
-                                                ch->msg_size);
+                       msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->
+                                                    remote_msgqueue + (get %
+                                                    ch->remote_nentries) *
+                                                    ch->entry_size);
 
-                       if (!(msg->flags & XPC_M_DONE))
+                       if (!(msg->flags & XPC_M_SN2_DONE))
                                break;
 
                        msg_flags |= msg->flags;
@@ -2251,11 +2255,11 @@ xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
                dev_dbg(xpc_chan, "local_GP->get changed to %ld, partid=%d, "
                        "channel=%d\n", get, ch->partid, ch->number);
 
-               send_msgrequest = (msg_flags & XPC_M_INTERRUPT);
+               send_msgrequest = (msg_flags & XPC_M_SN2_INTERRUPT);
 
                /*
                 * We need to ensure that the message referenced by
-                * local_GP->get is not XPC_M_DONE or that local_GP->get
+                * local_GP->get is not XPC_M_SN2_DONE or that local_GP->get
                 * equals w_local_GP.get, so we'll go have a look.
                 */
                initial_get = get;
@@ -2266,19 +2270,23 @@ xpc_acknowledge_msgs_sn2(struct xpc_channel *ch, s64 initial_get, u8 msg_flags)
 }
 
 static void
-xpc_received_msg_sn2(struct xpc_channel *ch, struct xpc_msg *msg)
+xpc_received_payload_sn2(struct xpc_channel *ch, void *payload)
 {
+       struct xpc_msg_sn2 *msg;
+       s64 msg_number;
        s64 get;
-       s64 msg_number = msg->number;
+
+       msg = container_of(payload, struct xpc_msg_sn2, payload);
+       msg_number = msg->number;
 
        dev_dbg(xpc_chan, "msg=0x%p, msg_number=%ld, partid=%d, channel=%d\n",
                (void *)msg, msg_number, ch->partid, ch->number);
 
-       DBUG_ON((((u64)msg - (u64)ch->remote_msgqueue) / ch->msg_size) !=
+       DBUG_ON((((u64)msg - (u64)ch->remote_msgqueue) / ch->entry_size) !=
                msg_number % ch->remote_nentries);
-       DBUG_ON(msg->flags & XPC_M_DONE);
+       DBUG_ON(msg->flags & XPC_M_SN2_DONE);
 
-       msg->flags |= XPC_M_DONE;
+       msg->flags |= XPC_M_SN2_DONE;
 
        /*
         * The preceding store of msg->flags must occur before the following
@@ -2337,8 +2345,8 @@ xpc_init_sn2(void)
 
        xpc_notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_sn2;
        xpc_process_msg_chctl_flags = xpc_process_msg_chctl_flags_sn2;
-       xpc_n_of_deliverable_msgs = xpc_n_of_deliverable_msgs_sn2;
-       xpc_get_deliverable_msg = xpc_get_deliverable_msg_sn2;
+       xpc_n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_sn2;
+       xpc_get_deliverable_payload = xpc_get_deliverable_payload_sn2;
 
        xpc_indicate_partition_engaged = xpc_indicate_partition_engaged_sn2;
        xpc_indicate_partition_disengaged =
@@ -2347,8 +2355,14 @@ xpc_init_sn2(void)
        xpc_any_partition_engaged = xpc_any_partition_engaged_sn2;
        xpc_assume_partition_disengaged = xpc_assume_partition_disengaged_sn2;
 
-       xpc_send_msg = xpc_send_msg_sn2;
-       xpc_received_msg = xpc_received_msg_sn2;
+       xpc_send_payload = xpc_send_payload_sn2;
+       xpc_received_payload = xpc_received_payload_sn2;
+
+       if (offsetof(struct xpc_msg_sn2, payload) > XPC_MSG_HDR_MAX_SIZE) {
+               dev_err(xpc_part, "header portion of struct xpc_msg_sn2 is "
+                       "larger than %d\n", XPC_MSG_HDR_MAX_SIZE);
+               return -E2BIG;
+       }
 
        buf_size = max(XPC_RP_VARS_SIZE,
                       XPC_RP_HEADER_SIZE + XP_NASID_MASK_BYTES_SN2);
index 689cb5c..1ac694c 100644 (file)
@@ -66,8 +66,11 @@ xpc_create_gru_mq_uv(unsigned int mq_size, int cpuid, unsigned int irq,
        mq_order = get_order(mq_size);
        page = alloc_pages_node(nid, GFP_KERNEL | __GFP_ZERO | GFP_THISNODE,
                                mq_order);
-       if (page == NULL)
+       if (page == NULL) {
+               dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
+                       "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
                return NULL;
+       }
 
        mq = page_address(page);
        ret = gru_create_message_queue(mq, mq_size);
@@ -193,202 +196,226 @@ xpc_process_activate_IRQ_rcvd_uv(void)
 
 }
 
-static irqreturn_t
-xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
+static void
+xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
+                             struct xpc_activate_mq_msghdr_uv *msg_hdr,
+                             int *wakeup_hb_checker)
 {
        unsigned long irq_flags;
-       struct xpc_activate_mq_msghdr_uv *msg_hdr;
-       short partid;
-       struct xpc_partition *part;
-       struct xpc_partition_uv *part_uv;
+       struct xpc_partition_uv *part_uv = &part->sn.uv;
        struct xpc_openclose_args *args;
-       int wakeup_hb_checker = 0;
 
-       while ((msg_hdr = gru_get_next_message(xpc_activate_mq_uv)) != NULL) {
+       part_uv->remote_act_state = msg_hdr->act_state;
 
-               partid = msg_hdr->partid;
-               if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
-                       dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() invalid"
-                               "partid=0x%x passed in message\n", partid);
-                       gru_free_message(xpc_activate_mq_uv, msg_hdr);
-                       continue;
-               }
-               part = &xpc_partitions[partid];
-               part_uv = &part->sn.uv;
+       switch (msg_hdr->type) {
+       case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
+               /* syncing of remote_act_state was just done above */
+               break;
 
-               part_uv->remote_act_state = msg_hdr->act_state;
+       case XPC_ACTIVATE_MQ_MSG_INC_HEARTBEAT_UV: {
+               struct xpc_activate_mq_msg_heartbeat_req_uv *msg;
 
-               switch (msg_hdr->type) {
-               case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
-                       /* syncing of remote_act_state was just done above */
-                       break;
+               msg = container_of(msg_hdr,
+                                  struct xpc_activate_mq_msg_heartbeat_req_uv,
+                                  hdr);
+               part_uv->heartbeat = msg->heartbeat;
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_OFFLINE_HEARTBEAT_UV: {
+               struct xpc_activate_mq_msg_heartbeat_req_uv *msg;
+
+               msg = container_of(msg_hdr,
+                                  struct xpc_activate_mq_msg_heartbeat_req_uv,
+                                  hdr);
+               part_uv->heartbeat = msg->heartbeat;
+
+               spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+               part_uv->flags |= XPC_P_HEARTBEAT_OFFLINE_UV;
+               spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_ONLINE_HEARTBEAT_UV: {
+               struct xpc_activate_mq_msg_heartbeat_req_uv *msg;
+
+               msg = container_of(msg_hdr,
+                                  struct xpc_activate_mq_msg_heartbeat_req_uv,
+                                  hdr);
+               part_uv->heartbeat = msg->heartbeat;
+
+               spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+               part_uv->flags &= ~XPC_P_HEARTBEAT_OFFLINE_UV;
+               spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
+               struct xpc_activate_mq_msg_activate_req_uv *msg;
 
-               case XPC_ACTIVATE_MQ_MSG_INC_HEARTBEAT_UV: {
-                       struct xpc_activate_mq_msg_heartbeat_req_uv *msg;
+               /*
+                * ??? Do we deal here with ts_jiffies being different
+                * ??? if act_state != XPC_P_AS_INACTIVE instead of
+                * ??? below?
+                */
+               msg = container_of(msg_hdr, struct
+                                  xpc_activate_mq_msg_activate_req_uv, hdr);
 
-                       msg = (struct xpc_activate_mq_msg_heartbeat_req_uv *)
-                           msg_hdr;
-                       part_uv->heartbeat = msg->heartbeat;
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_OFFLINE_HEARTBEAT_UV: {
-                       struct xpc_activate_mq_msg_heartbeat_req_uv *msg;
-
-                       msg = (struct xpc_activate_mq_msg_heartbeat_req_uv *)
-                           msg_hdr;
-                       part_uv->heartbeat = msg->heartbeat;
-                       spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
-                       part_uv->flags |= XPC_P_HEARTBEAT_OFFLINE_UV;
-                       spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_ONLINE_HEARTBEAT_UV: {
-                       struct xpc_activate_mq_msg_heartbeat_req_uv *msg;
-
-                       msg = (struct xpc_activate_mq_msg_heartbeat_req_uv *)
-                           msg_hdr;
-                       part_uv->heartbeat = msg->heartbeat;
-                       spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
-                       part_uv->flags &= ~XPC_P_HEARTBEAT_OFFLINE_UV;
-                       spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
-                       struct xpc_activate_mq_msg_activate_req_uv *msg;
-
-                       /*
-                        * ??? Do we deal here with ts_jiffies being different
-                        * ??? if act_state != XPC_P_AS_INACTIVE instead of
-                        * ??? below?
-                        */
-                       msg = (struct xpc_activate_mq_msg_activate_req_uv *)
-                           msg_hdr;
-                       spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock,
-                                         irq_flags);
-                       if (part_uv->act_state_req == 0)
-                               xpc_activate_IRQ_rcvd++;
-                       part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
-                       part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
-                       part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
-                       part_uv->remote_activate_mq_gpa = msg->activate_mq_gpa;
-                       spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock,
-                                              irq_flags);
-                       wakeup_hb_checker++;
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
-                       struct xpc_activate_mq_msg_deactivate_req_uv *msg;
-
-                       msg = (struct xpc_activate_mq_msg_deactivate_req_uv *)
-                           msg_hdr;
-                       spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock,
-                                         irq_flags);
-                       if (part_uv->act_state_req == 0)
-                               xpc_activate_IRQ_rcvd++;
-                       part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
-                       part_uv->reason = msg->reason;
-                       spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock,
-                                              irq_flags);
-                       wakeup_hb_checker++;
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
-                       struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
+               spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+               if (part_uv->act_state_req == 0)
+                       xpc_activate_IRQ_rcvd++;
+               part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
+               part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
+               part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
+               part_uv->remote_activate_mq_gpa = msg->activate_mq_gpa;
+               spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 
-                       msg = (struct xpc_activate_mq_msg_chctl_closerequest_uv
-                           *)msg_hdr;
-                       args = &part->remote_openclose_args[msg->ch_number];
-                       args->reason = msg->reason;
+               (*wakeup_hb_checker)++;
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
+               struct xpc_activate_mq_msg_deactivate_req_uv *msg;
 
-                       spin_lock_irqsave(&part->chctl_lock, irq_flags);
-                       part->chctl.flags[msg->ch_number] |=
-                           XPC_CHCTL_CLOSEREQUEST;
-                       spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
+               msg = container_of(msg_hdr, struct
+                                  xpc_activate_mq_msg_deactivate_req_uv, hdr);
 
-                       xpc_wakeup_channel_mgr(part);
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
-                       struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
+               spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+               if (part_uv->act_state_req == 0)
+                       xpc_activate_IRQ_rcvd++;
+               part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
+               part_uv->reason = msg->reason;
+               spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+
+               (*wakeup_hb_checker)++;
+               return;
+       }
+       case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
+               struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
 
-                       msg = (struct xpc_activate_mq_msg_chctl_closereply_uv *)
-                           msg_hdr;
+               msg = container_of(msg_hdr, struct
+                                  xpc_activate_mq_msg_chctl_closerequest_uv,
+                                  hdr);
+               args = &part->remote_openclose_args[msg->ch_number];
+               args->reason = msg->reason;
 
-                       spin_lock_irqsave(&part->chctl_lock, irq_flags);
-                       part->chctl.flags[msg->ch_number] |=
-                           XPC_CHCTL_CLOSEREPLY;
-                       spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
+               spin_lock_irqsave(&part->chctl_lock, irq_flags);
+               part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
+               spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 
-                       xpc_wakeup_channel_mgr(part);
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
-                       struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
+               xpc_wakeup_channel_mgr(part);
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
+               struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
 
-                       msg = (struct xpc_activate_mq_msg_chctl_openrequest_uv
-                           *)msg_hdr;
-                       args = &part->remote_openclose_args[msg->ch_number];
-                       args->msg_size = msg->msg_size;
-                       args->local_nentries = msg->local_nentries;
+               msg = container_of(msg_hdr, struct
+                                  xpc_activate_mq_msg_chctl_closereply_uv,
+                                  hdr);
 
-                       spin_lock_irqsave(&part->chctl_lock, irq_flags);
-                       part->chctl.flags[msg->ch_number] |=
-                           XPC_CHCTL_OPENREQUEST;
-                       spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
+               spin_lock_irqsave(&part->chctl_lock, irq_flags);
+               part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
+               spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
 
-                       xpc_wakeup_channel_mgr(part);
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
-                       struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
-
-                       msg = (struct xpc_activate_mq_msg_chctl_openreply_uv *)
-                           msg_hdr;
-                       args = &part->remote_openclose_args[msg->ch_number];
-                       args->remote_nentries = msg->remote_nentries;
-                       args->local_nentries = msg->local_nentries;
-                       args->local_msgqueue_pa = msg->local_notify_mq_gpa;
-
-                       spin_lock_irqsave(&part->chctl_lock, irq_flags);
-                       part->chctl.flags[msg->ch_number] |=
-                           XPC_CHCTL_OPENREPLY;
-                       spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
-
-                       xpc_wakeup_channel_mgr(part);
-                       break;
-               }
-               case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
-                       spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
-                       part_uv->flags |= XPC_P_ENGAGED_UV;
-                       spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
-                       break;
+               xpc_wakeup_channel_mgr(part);
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
+               struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
+
+               msg = container_of(msg_hdr, struct
+                                  xpc_activate_mq_msg_chctl_openrequest_uv,
+                                  hdr);
+               args = &part->remote_openclose_args[msg->ch_number];
+               args->entry_size = msg->entry_size;
+               args->local_nentries = msg->local_nentries;
+
+               spin_lock_irqsave(&part->chctl_lock, irq_flags);
+               part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
+               spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
+
+               xpc_wakeup_channel_mgr(part);
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
+               struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
+
+               msg = container_of(msg_hdr, struct
+                                  xpc_activate_mq_msg_chctl_openreply_uv, hdr);
+               args = &part->remote_openclose_args[msg->ch_number];
+               args->remote_nentries = msg->remote_nentries;
+               args->local_nentries = msg->local_nentries;
+               args->local_msgqueue_pa = msg->local_notify_mq_gpa;
+
+               spin_lock_irqsave(&part->chctl_lock, irq_flags);
+               part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
+               spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
+
+               xpc_wakeup_channel_mgr(part);
+               break;
+       }
+       case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
+               spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+               part_uv->flags |= XPC_P_ENGAGED_UV;
+               spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+               break;
+
+       case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
+               spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+               part_uv->flags &= ~XPC_P_ENGAGED_UV;
+               spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+               break;
+
+       default:
+               dev_err(xpc_part, "received unknown activate_mq msg type=%d "
+                       "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
+
+               /* get hb checker to deactivate from the remote partition */
+               spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+               if (part_uv->act_state_req == 0)
+                       xpc_activate_IRQ_rcvd++;
+               part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
+               part_uv->reason = xpBadMsgType;
+               spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 
-               case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
-                       spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
-                       part_uv->flags &= ~XPC_P_ENGAGED_UV;
-                       spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
-                       break;
+               (*wakeup_hb_checker)++;
+               return;
+       }
 
-               default:
-                       dev_err(xpc_part, "received unknown activate_mq msg "
-                               "type=%d from partition=%d\n", msg_hdr->type,
-                               partid);
-               }
+       if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
+           part->remote_rp_ts_jiffies != 0) {
+               /*
+                * ??? Does what we do here need to be sensitive to
+                * ??? act_state or remote_act_state?
+                */
+               spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+               if (part_uv->act_state_req == 0)
+                       xpc_activate_IRQ_rcvd++;
+               part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
+               spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
 
-               if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
-                   part->remote_rp_ts_jiffies != 0) {
-                       /*
-                        * ??? Does what we do here need to be sensitive to
-                        * ??? act_state or remote_act_state?
-                        */
-                       spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock,
-                                         irq_flags);
-                       if (part_uv->act_state_req == 0)
-                               xpc_activate_IRQ_rcvd++;
-                       part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
-                       spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock,
-                                              irq_flags);
-                       wakeup_hb_checker++;
+               (*wakeup_hb_checker)++;
+       }
+}
+
+static irqreturn_t
+xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
+{
+       struct xpc_activate_mq_msghdr_uv *msg_hdr;
+       short partid;
+       struct xpc_partition *part;
+       int wakeup_hb_checker = 0;
+
+       while ((msg_hdr = gru_get_next_message(xpc_activate_mq_uv)) != NULL) {
+
+               partid = msg_hdr->partid;
+               if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
+                       dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
+                               "received invalid partid=0x%x in message\n",
+                               partid);
+               } else {
+                       part = &xpc_partitions[partid];
+                       if (xpc_part_ref(part)) {
+                               xpc_handle_activate_mq_msg_uv(part, msg_hdr,
+                                                           &wakeup_hb_checker);
+                               xpc_part_deref(part);
+                       }
                }
 
                gru_free_message(xpc_activate_mq_uv, msg_hdr);
@@ -616,14 +643,82 @@ xpc_request_partition_deactivation_uv(struct xpc_partition *part)
        }
 }
 
+static void
+xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
+{
+       /* nothing needs to be done */
+       return;
+}
+
+static void
+xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
+{
+       head->first = NULL;
+       head->last = NULL;
+       spin_lock_init(&head->lock);
+       head->n_entries = 0;
+}
+
+static void *
+xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
+{
+       unsigned long irq_flags;
+       struct xpc_fifo_entry_uv *first;
+
+       spin_lock_irqsave(&head->lock, irq_flags);
+       first = head->first;
+       if (head->first != NULL) {
+               head->first = first->next;
+               if (head->first == NULL)
+                       head->last = NULL;
+       }
+       head->n_entries++;
+       spin_unlock_irqrestore(&head->lock, irq_flags);
+       first->next = NULL;
+       return first;
+}
+
+static void
+xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
+                     struct xpc_fifo_entry_uv *last)
+{
+       unsigned long irq_flags;
+
+       last->next = NULL;
+       spin_lock_irqsave(&head->lock, irq_flags);
+       if (head->last != NULL)
+               head->last->next = last;
+       else
+               head->first = last;
+       head->last = last;
+       head->n_entries--;
+       BUG_ON(head->n_entries < 0);
+       spin_unlock_irqrestore(&head->lock, irq_flags);
+}
+
+static int
+xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
+{
+       return head->n_entries;
+}
+
 /*
  * Setup the channel structures that are uv specific.
  */
 static enum xp_retval
 xpc_setup_ch_structures_sn_uv(struct xpc_partition *part)
 {
-       /* !!! this function needs fleshing out */
-       return xpUnsupported;
+       struct xpc_channel_uv *ch_uv;
+       int ch_number;
+
+       for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
+               ch_uv = &part->channels[ch_number].sn.uv;
+
+               xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
+               xpc_init_fifo_uv(&ch_uv->recv_msg_list);
+       }
+
+       return xpSuccess;
 }
 
 /*
@@ -632,7 +727,7 @@ xpc_setup_ch_structures_sn_uv(struct xpc_partition *part)
 static void
 xpc_teardown_ch_structures_sn_uv(struct xpc_partition *part)
 {
-       /* !!! this function needs fleshing out */
+       /* nothing needs to be done */
        return;
 }
 
@@ -680,20 +775,114 @@ xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
 }
 
 static enum xp_retval
+xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
+{
+       struct xpc_channel_uv *ch_uv = &ch->sn.uv;
+       struct xpc_send_msg_slot_uv *msg_slot;
+       unsigned long irq_flags;
+       int nentries;
+       int entry;
+       size_t nbytes;
+
+       for (nentries = ch->local_nentries; nentries > 0; nentries--) {
+               nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
+               ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
+               if (ch_uv->send_msg_slots == NULL)
+                       continue;
+
+               for (entry = 0; entry < nentries; entry++) {
+                       msg_slot = &ch_uv->send_msg_slots[entry];
+
+                       msg_slot->msg_slot_number = entry;
+                       xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
+                                             &msg_slot->next);
+               }
+
+               spin_lock_irqsave(&ch->lock, irq_flags);
+               if (nentries < ch->local_nentries)
+                       ch->local_nentries = nentries;
+               spin_unlock_irqrestore(&ch->lock, irq_flags);
+               return xpSuccess;
+       }
+
+       return xpNoMemory;
+}
+
+static enum xp_retval
+xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
+{
+       struct xpc_channel_uv *ch_uv = &ch->sn.uv;
+       struct xpc_notify_mq_msg_uv *msg_slot;
+       unsigned long irq_flags;
+       int nentries;
+       int entry;
+       size_t nbytes;
+
+       for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
+               nbytes = nentries * ch->entry_size;
+               ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
+               if (ch_uv->recv_msg_slots == NULL)
+                       continue;
+
+               for (entry = 0; entry < nentries; entry++) {
+                       msg_slot = ch_uv->recv_msg_slots + entry *
+                           ch->entry_size;
+
+                       msg_slot->hdr.msg_slot_number = entry;
+               }
+
+               spin_lock_irqsave(&ch->lock, irq_flags);
+               if (nentries < ch->remote_nentries)
+                       ch->remote_nentries = nentries;
+               spin_unlock_irqrestore(&ch->lock, irq_flags);
+               return xpSuccess;
+       }
+
+       return xpNoMemory;
+}
+
+/*
+ * Allocate msg_slots associated with the channel.
+ */
+static enum xp_retval
 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
 {
-       /* !!! this function needs fleshing out */
-       return xpUnsupported;
+       static enum xp_retval ret;
+       struct xpc_channel_uv *ch_uv = &ch->sn.uv;
+
+       DBUG_ON(ch->flags & XPC_C_SETUP);
+
+       ret = xpc_allocate_send_msg_slot_uv(ch);
+       if (ret == xpSuccess) {
+
+               ret = xpc_allocate_recv_msg_slot_uv(ch);
+               if (ret != xpSuccess) {
+                       kfree(ch_uv->send_msg_slots);
+                       xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
+               }
+       }
+       return ret;
 }
 
+/*
+ * Free up msg_slots and clear other stuff that were setup for the specified
+ * channel.
+ */
 static void
 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
 {
        struct xpc_channel_uv *ch_uv = &ch->sn.uv;
 
+       DBUG_ON(!spin_is_locked(&ch->lock));
+
        ch_uv->remote_notify_mq_gpa = 0;
 
-       /* !!! this function needs fleshing out */
+       if (ch->flags & XPC_C_SETUP) {
+               xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
+               kfree(ch_uv->send_msg_slots);
+               xpc_init_fifo_uv(&ch_uv->recv_msg_list);
+               kfree(ch_uv->recv_msg_slots);
+       }
 }
 
 static void
@@ -723,7 +912,7 @@ xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
        struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
 
        msg.ch_number = ch->number;
-       msg.msg_size = ch->msg_size;
+       msg.entry_size = ch->entry_size;
        msg.local_nentries = ch->local_nentries;
        xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
                                    XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
@@ -743,6 +932,18 @@ xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
 }
 
 static void
+xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
+{
+       unsigned long irq_flags;
+
+       spin_lock_irqsave(&part->chctl_lock, irq_flags);
+       part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
+       spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
+
+       xpc_wakeup_channel_mgr(part);
+}
+
+static void
 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
                               unsigned long msgqueue_pa)
 {
@@ -798,11 +999,358 @@ xpc_any_partition_engaged_uv(void)
        return 0;
 }
 
-static struct xpc_msg *
-xpc_get_deliverable_msg_uv(struct xpc_channel *ch)
+static enum xp_retval
+xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
+                        struct xpc_send_msg_slot_uv **address_of_msg_slot)
+{
+       enum xp_retval ret;
+       struct xpc_send_msg_slot_uv *msg_slot;
+       struct xpc_fifo_entry_uv *entry;
+
+       while (1) {
+               entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
+               if (entry != NULL)
+                       break;
+
+               if (flags & XPC_NOWAIT)
+                       return xpNoWait;
+
+               ret = xpc_allocate_msg_wait(ch);
+               if (ret != xpInterrupted && ret != xpTimeout)
+                       return ret;
+       }
+
+       msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
+       *address_of_msg_slot = msg_slot;
+       return xpSuccess;
+}
+
+static void
+xpc_free_msg_slot_uv(struct xpc_channel *ch,
+                    struct xpc_send_msg_slot_uv *msg_slot)
+{
+       xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
+
+       /* wakeup anyone waiting for a free msg slot */
+       if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
+               wake_up(&ch->msg_allocate_wq);
+}
+
+static void
+xpc_notify_sender_uv(struct xpc_channel *ch,
+                    struct xpc_send_msg_slot_uv *msg_slot,
+                    enum xp_retval reason)
+{
+       xpc_notify_func func = msg_slot->func;
+
+       if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
+
+               atomic_dec(&ch->n_to_notify);
+
+               dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
+                       "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
+                       msg_slot->msg_slot_number, ch->partid, ch->number);
+
+               func(reason, ch->partid, ch->number, msg_slot->key);
+
+               dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
+                       "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
+                       msg_slot->msg_slot_number, ch->partid, ch->number);
+       }
+}
+
+static void
+xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
+                           struct xpc_notify_mq_msg_uv *msg)
+{
+       struct xpc_send_msg_slot_uv *msg_slot;
+       int entry = msg->hdr.msg_slot_number % ch->local_nentries;
+
+       msg_slot = &ch->sn.uv.send_msg_slots[entry];
+
+       BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
+       msg_slot->msg_slot_number += ch->local_nentries;
+
+       if (msg_slot->func != NULL)
+               xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
+
+       xpc_free_msg_slot_uv(ch, msg_slot);
+}
+
+static void
+xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
+                           struct xpc_notify_mq_msg_uv *msg)
+{
+       struct xpc_partition_uv *part_uv = &part->sn.uv;
+       struct xpc_channel *ch;
+       struct xpc_channel_uv *ch_uv;
+       struct xpc_notify_mq_msg_uv *msg_slot;
+       unsigned long irq_flags;
+       int ch_number = msg->hdr.ch_number;
+
+       if (unlikely(ch_number >= part->nchannels)) {
+               dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
+                       "channel number=0x%x in message from partid=%d\n",
+                       ch_number, XPC_PARTID(part));
+
+               /* get hb checker to deactivate from the remote partition */
+               spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+               if (part_uv->act_state_req == 0)
+                       xpc_activate_IRQ_rcvd++;
+               part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
+               part_uv->reason = xpBadChannelNumber;
+               spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
+
+               wake_up_interruptible(&xpc_activate_IRQ_wq);
+               return;
+       }
+
+       ch = &part->channels[ch_number];
+       xpc_msgqueue_ref(ch);
+
+       if (!(ch->flags & XPC_C_CONNECTED)) {
+               xpc_msgqueue_deref(ch);
+               return;
+       }
+
+       /* see if we're really dealing with an ACK for a previously sent msg */
+       if (msg->hdr.size == 0) {
+               xpc_handle_notify_mq_ack_uv(ch, msg);
+               xpc_msgqueue_deref(ch);
+               return;
+       }
+
+       /* we're dealing with a normal message sent via the notify_mq */
+       ch_uv = &ch->sn.uv;
+
+       msg_slot = (struct xpc_notify_mq_msg_uv *)((u64)ch_uv->recv_msg_slots +
+                   (msg->hdr.msg_slot_number % ch->remote_nentries) *
+                   ch->entry_size);
+
+       BUG_ON(msg->hdr.msg_slot_number != msg_slot->hdr.msg_slot_number);
+       BUG_ON(msg_slot->hdr.size != 0);
+
+       memcpy(msg_slot, msg, msg->hdr.size);
+
+       xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
+
+       if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
+               /*
+                * If there is an existing idle kthread get it to deliver
+                * the payload, otherwise we'll have to get the channel mgr
+                * for this partition to create a kthread to do the delivery.
+                */
+               if (atomic_read(&ch->kthreads_idle) > 0)
+                       wake_up_nr(&ch->idle_wq, 1);
+               else
+                       xpc_send_chctl_local_msgrequest_uv(part, ch->number);
+       }
+       xpc_msgqueue_deref(ch);
+}
+
+static irqreturn_t
+xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
+{
+       struct xpc_notify_mq_msg_uv *msg;
+       short partid;
+       struct xpc_partition *part;
+
+       while ((msg = gru_get_next_message(xpc_notify_mq_uv)) != NULL) {
+
+               partid = msg->hdr.partid;
+               if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
+                       dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
+                               "invalid partid=0x%x in message\n", partid);
+               } else {
+                       part = &xpc_partitions[partid];
+
+                       if (xpc_part_ref(part)) {
+                               xpc_handle_notify_mq_msg_uv(part, msg);
+                               xpc_part_deref(part);
+                       }
+               }
+
+               gru_free_message(xpc_notify_mq_uv, msg);
+       }
+
+       return IRQ_HANDLED;
+}
+
+static int
+xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
+{
+       return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
+}
+
+static void
+xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
+{
+       struct xpc_channel *ch = &part->channels[ch_number];
+       int ndeliverable_payloads;
+
+       xpc_msgqueue_ref(ch);
+
+       ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
+
+       if (ndeliverable_payloads > 0 &&
+           (ch->flags & XPC_C_CONNECTED) &&
+           (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
+
+               xpc_activate_kthreads(ch, ndeliverable_payloads);
+       }
+
+       xpc_msgqueue_deref(ch);
+}
+
+static enum xp_retval
+xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
+                   u16 payload_size, u8 notify_type, xpc_notify_func func,
+                   void *key)
+{
+       enum xp_retval ret = xpSuccess;
+       struct xpc_send_msg_slot_uv *msg_slot = NULL;
+       struct xpc_notify_mq_msg_uv *msg;
+       u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
+       size_t msg_size;
+
+       DBUG_ON(notify_type != XPC_N_CALL);
+
+       msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
+       if (msg_size > ch->entry_size)
+               return xpPayloadTooBig;
+
+       xpc_msgqueue_ref(ch);
+
+       if (ch->flags & XPC_C_DISCONNECTING) {
+               ret = ch->reason;
+               goto out_1;
+       }
+       if (!(ch->flags & XPC_C_CONNECTED)) {
+               ret = xpNotConnected;
+               goto out_1;
+       }
+
+       ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
+       if (ret != xpSuccess)
+               goto out_1;
+
+       if (func != NULL) {
+               atomic_inc(&ch->n_to_notify);
+
+               msg_slot->key = key;
+               wmb(); /* a non-NULL func must hit memory after the key */
+               msg_slot->func = func;
+
+               if (ch->flags & XPC_C_DISCONNECTING) {
+                       ret = ch->reason;
+                       goto out_2;
+               }
+       }
+
+       msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
+       msg->hdr.partid = xp_partition_id;
+       msg->hdr.ch_number = ch->number;
+       msg->hdr.size = msg_size;
+       msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
+       memcpy(&msg->payload, payload, payload_size);
+
+       ret = xpc_send_gru_msg(ch->sn.uv.remote_notify_mq_gpa, msg, msg_size);
+       if (ret == xpSuccess)
+               goto out_1;
+
+       XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
+out_2:
+       if (func != NULL) {
+               /*
+                * Try to NULL the msg_slot's func field. If we fail, then
+                * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
+                * case we need to pretend we succeeded to send the message
+                * since the user will get a callout for the disconnect error
+                * by xpc_notify_senders_of_disconnect_uv(), and to also get an
+                * error returned here will confuse them. Additionally, since
+                * in this case the channel is being disconnected we don't need
+                * to put the the msg_slot back on the free list.
+                */
+               if (cmpxchg(&msg_slot->func, func, NULL) != func) {
+                       ret = xpSuccess;
+                       goto out_1;
+               }
+
+               msg_slot->key = NULL;
+               atomic_dec(&ch->n_to_notify);
+       }
+       xpc_free_msg_slot_uv(ch, msg_slot);
+out_1:
+       xpc_msgqueue_deref(ch);
+       return ret;
+}
+
+/*
+ * Tell the callers of xpc_send_notify() that the status of their payloads
+ * is unknown because the channel is now disconnecting.
+ *
+ * We don't worry about putting these msg_slots on the free list since the
+ * msg_slots themselves are about to be kfree'd.
+ */
+static void
+xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
+{
+       struct xpc_send_msg_slot_uv *msg_slot;
+       int entry;
+
+       DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
+
+       for (entry = 0; entry < ch->local_nentries; entry++) {
+
+               if (atomic_read(&ch->n_to_notify) == 0)
+                       break;
+
+               msg_slot = &ch->sn.uv.send_msg_slots[entry];
+               if (msg_slot->func != NULL)
+                       xpc_notify_sender_uv(ch, msg_slot, ch->reason);
+       }
+}
+
+/*
+ * Get the next deliverable message's payload.
+ */
+static void *
+xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
+{
+       struct xpc_fifo_entry_uv *entry;
+       struct xpc_notify_mq_msg_uv *msg;
+       void *payload = NULL;
+
+       if (!(ch->flags & XPC_C_DISCONNECTING)) {
+               entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
+               if (entry != NULL) {
+                       msg = container_of(entry, struct xpc_notify_mq_msg_uv,
+                                          hdr.u.next);
+                       payload = &msg->payload;
+               }
+       }
+       return payload;
+}
+
+static void
+xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
 {
-       /* !!! this function needs fleshing out */
-       return NULL;
+       struct xpc_notify_mq_msg_uv *msg;
+       enum xp_retval ret;
+
+       msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
+
+       /* return an ACK to the sender of this message */
+
+       msg->hdr.partid = xp_partition_id;
+       msg->hdr.size = 0;      /* size of zero indicates this is an ACK */
+
+       ret = xpc_send_gru_msg(ch->sn.uv.remote_notify_mq_gpa, msg,
+                              sizeof(struct xpc_notify_mq_msghdr_uv));
+       if (ret != xpSuccess)
+               XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
+
+       msg->hdr.msg_slot_number += ch->remote_nentries;
 }
 
 int
@@ -824,6 +1372,8 @@ xpc_init_uv(void)
            xpc_request_partition_reactivation_uv;
        xpc_request_partition_deactivation =
            xpc_request_partition_deactivation_uv;
+       xpc_cancel_partition_deactivation_request =
+           xpc_cancel_partition_deactivation_request_uv;
 
        xpc_setup_ch_structures_sn = xpc_setup_ch_structures_sn_uv;
        xpc_teardown_ch_structures_sn = xpc_teardown_ch_structures_sn_uv;
@@ -848,7 +1398,18 @@ xpc_init_uv(void)
        xpc_partition_engaged = xpc_partition_engaged_uv;
        xpc_any_partition_engaged = xpc_any_partition_engaged_uv;
 
-       xpc_get_deliverable_msg = xpc_get_deliverable_msg_uv;
+       xpc_n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv;
+       xpc_process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv;
+       xpc_send_payload = xpc_send_payload_uv;
+       xpc_notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv;
+       xpc_get_deliverable_payload = xpc_get_deliverable_payload_uv;
+       xpc_received_payload = xpc_received_payload_uv;
+
+       if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
+               dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
+                       XPC_MSG_HDR_MAX_SIZE);
+               return -E2BIG;
+       }
 
        /* ??? The cpuid argument's value is 0, is that what we want? */
        /* !!! The irq argument's value isn't correct. */
@@ -857,6 +1418,17 @@ xpc_init_uv(void)
        if (xpc_activate_mq_uv == NULL)
                return -ENOMEM;
 
+       /* ??? The cpuid argument's value is 0, is that what we want? */
+       /* !!! The irq argument's value isn't correct. */
+       xpc_notify_mq_uv = xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, 0, 0,
+                                               xpc_handle_notify_IRQ_uv);
+       if (xpc_notify_mq_uv == NULL) {
+               /* !!! The irq argument's value isn't correct. */
+               xpc_destroy_gru_mq_uv(xpc_activate_mq_uv,
+                                     XPC_ACTIVATE_MQ_SIZE_UV, 0);
+               return -ENOMEM;
+       }
+
        return 0;
 }
 
@@ -864,5 +1436,8 @@ void
 xpc_exit_uv(void)
 {
        /* !!! The irq argument's value isn't correct. */
+       xpc_destroy_gru_mq_uv(xpc_notify_mq_uv, XPC_NOTIFY_MQ_SIZE_UV, 0);
+
+       /* !!! The irq argument's value isn't correct. */
        xpc_destroy_gru_mq_uv(xpc_activate_mq_uv, XPC_ACTIVATE_MQ_SIZE_UV, 0);
 }
index 4f5d622..71513b3 100644 (file)
@@ -57,11 +57,10 @@ struct xpnet_message {
  *
  * XPC expects each message to exist in an individual cacheline.
  */
-#define XPNET_MSG_SIZE         (L1_CACHE_BYTES - XPC_MSG_PAYLOAD_OFFSET)
+#define XPNET_MSG_SIZE         XPC_MSG_PAYLOAD_MAX_SIZE
 #define XPNET_MSG_DATA_MAX     \
-               (XPNET_MSG_SIZE - (u64)(&((struct xpnet_message *)0)->data))
-#define XPNET_MSG_ALIGNED_SIZE (L1_CACHE_ALIGN(XPNET_MSG_SIZE))
-#define XPNET_MSG_NENTRIES     (PAGE_SIZE / XPNET_MSG_ALIGNED_SIZE)
+               (XPNET_MSG_SIZE - offsetof(struct xpnet_message, data))
+#define XPNET_MSG_NENTRIES     (PAGE_SIZE / XPC_MSG_MAX_SIZE)
 
 #define XPNET_MAX_KTHREADS     (XPNET_MSG_NENTRIES + 1)
 #define XPNET_MAX_IDLE_KTHREADS        (XPNET_MSG_NENTRIES + 1)
@@ -408,6 +407,7 @@ xpnet_send(struct sk_buff *skb, struct xpnet_pending_msg *queued_msg,
 {
        u8 msg_buffer[XPNET_MSG_SIZE];
        struct xpnet_message *msg = (struct xpnet_message *)&msg_buffer;
+       u16 msg_size = sizeof(struct xpnet_message);
        enum xp_retval ret;
 
        msg->embedded_bytes = embedded_bytes;
@@ -417,6 +417,7 @@ xpnet_send(struct sk_buff *skb, struct xpnet_pending_msg *queued_msg,
                        &msg->data, skb->data, (size_t)embedded_bytes);
                skb_copy_from_linear_data(skb, &msg->data,
                                          (size_t)embedded_bytes);
+               msg_size += embedded_bytes - 1;
        } else {
                msg->version = XPNET_VERSION;
        }
@@ -435,7 +436,7 @@ xpnet_send(struct sk_buff *skb, struct xpnet_pending_msg *queued_msg,
        atomic_inc(&queued_msg->use_count);
 
        ret = xpc_send_notify(dest_partid, XPC_NET_CHANNEL, XPC_NOWAIT, msg,
-                             XPNET_MSG_SIZE, xpnet_send_completed, queued_msg);
+                             msg_size, xpnet_send_completed, queued_msg);
        if (unlikely(ret != xpSuccess))
                atomic_dec(&queued_msg->use_count);
 }