#define XPC_NASID_FROM_W_B(_w, _b) (((_w) * 64 + (_b)) * 2)
#define XPC_HB_DEFAULT_INTERVAL 5 /* incr HB every x secs */
-#define XPC_HB_CHECK_DEFAULT_TIMEOUT 20 /* check HB every x secs */
+#define XPC_HB_CHECK_DEFAULT_INTERVAL 20 /* check HB every x secs */
/* define the process name of HB checker and the CPU it is pinned to */
#define XPC_HB_CHECK_THREAD_NAME "xpc_hb"
#define XPC_DISCOVERY_THREAD_NAME "xpc_discovery"
-#define XPC_HB_ALLOWED(_p, _v) ((_v)->heartbeating_to_mask & (1UL << (_p)))
-#define XPC_ALLOW_HB(_p, _v) (_v)->heartbeating_to_mask |= (1UL << (_p))
-#define XPC_DISALLOW_HB(_p, _v) (_v)->heartbeating_to_mask &= (~(1UL << (_p)))
-
-
/*
* Reserved Page provided by SAL.
*
u8 version;
u8 pad[6]; /* pad to u64 align */
volatile u64 vars_pa;
+ struct timespec stamp; /* time when reserved page was initialized */
u64 part_nasids[XP_NASID_MASK_WORDS] ____cacheline_aligned;
u64 mach_nasids[XP_NASID_MASK_WORDS] ____cacheline_aligned;
};
-#define XPC_RP_VERSION _XPC_VERSION(1,0) /* version 1.0 of the reserved page */
#define XPC_RSVD_PAGE_ALIGNED_SIZE \
(L1_CACHE_ALIGN(sizeof(struct xpc_rsvd_page)))
+#define XPC_RP_VERSION _XPC_VERSION(1,1) /* version 1.1 of the reserved page */
+
+#define XPC_SUPPORTS_RP_STAMP(_version) \
+ (_version >= _XPC_VERSION(1,1))
+
+/*
+ * compare stamps - the return value is:
+ *
+ * < 0, if stamp1 < stamp2
+ * = 0, if stamp1 == stamp2
+ * > 0, if stamp1 > stamp2
+ */
+static inline int
+xpc_compare_stamps(struct timespec *stamp1, struct timespec *stamp2)
+{
+ int ret;
+
+
+ if ((ret = stamp1->tv_sec - stamp2->tv_sec) == 0) {
+ ret = stamp1->tv_nsec - stamp2->tv_nsec;
+ }
+ return ret;
+}
+
/*
* Define the structures by which XPC variables can be exported to other
u64 vars_part_pa;
u64 amos_page_pa; /* paddr of page of AMOs from MSPEC driver */
AMO_t *amos_page; /* vaddr of page of AMOs from MSPEC driver */
- AMO_t *act_amos; /* pointer to the first activation AMO */
};
-#define XPC_V_VERSION _XPC_VERSION(3,0) /* version 3.0 of the cross vars */
#define XPC_VARS_ALIGNED_SIZE (L1_CACHE_ALIGN(sizeof(struct xpc_vars)))
+#define XPC_V_VERSION _XPC_VERSION(3,1) /* version 3.1 of the cross vars */
+
+#define XPC_SUPPORTS_DISENGAGE_REQUEST(_version) \
+ (_version >= _XPC_VERSION(3,1))
+
+
+static inline int
+xpc_hb_allowed(partid_t partid, struct xpc_vars *vars)
+{
+ return ((vars->heartbeating_to_mask & (1UL << partid)) != 0);
+}
+
+static inline void
+xpc_allow_hb(partid_t partid, struct xpc_vars *vars)
+{
+ u64 old_mask, new_mask;
+
+ do {
+ old_mask = vars->heartbeating_to_mask;
+ new_mask = (old_mask | (1UL << partid));
+ } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
+ old_mask);
+}
+
+static inline void
+xpc_disallow_hb(partid_t partid, struct xpc_vars *vars)
+{
+ u64 old_mask, new_mask;
+
+ do {
+ old_mask = vars->heartbeating_to_mask;
+ new_mask = (old_mask & ~(1UL << partid));
+ } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
+ old_mask);
+}
+
+
+/*
+ * The AMOs page consists of a number of AMO variables which are divided into
+ * four groups, The first two groups are used to identify an IRQ's sender.
+ * These two groups consist of 64 and 16 AMO variables respectively. The last
+ * two groups, consisting of just one AMO variable each, are used to identify
+ * the remote partitions that are currently engaged (from the viewpoint of
+ * the XPC running on the remote partition).
+ */
+#define XPC_NOTIFY_IRQ_AMOS 0
+#define XPC_ACTIVATE_IRQ_AMOS (XPC_NOTIFY_IRQ_AMOS + XP_MAX_PARTITIONS)
+#define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
+#define XPC_DISENGAGE_REQUEST_AMO (XPC_ENGAGED_PARTITIONS_AMO + 1)
+
+
/*
* The following structure describes the per partition specific variables.
*
void *key; /* pointer to user's key */
struct semaphore msg_to_pull_sema; /* next msg to pull serialization */
- struct semaphore teardown_sema; /* wait for teardown completion */
+ struct semaphore wdisconnect_sema; /* wait for channel disconnect */
struct xpc_openclose_args *local_openclose_args; /* args passed on */
/* opening or closing of channel */
#define XPC_C_DISCONNECTED 0x00002000 /* channel is disconnected */
#define XPC_C_DISCONNECTING 0x00004000 /* channel is being disconnected */
+#define XPC_C_WDISCONNECT 0x00008000 /* waiting for channel disconnect */
/* XPC HB infrastructure */
+ u8 remote_rp_version; /* version# of partition's rsvd pg */
+ struct timespec remote_rp_stamp;/* time when rsvd pg was initialized */
u64 remote_rp_pa; /* phys addr of partition's rsvd pg */
u64 remote_vars_pa; /* phys addr of partition's vars */
u64 remote_vars_part_pa; /* phys addr of partition's vars part */
u32 act_IRQ_rcvd; /* IRQs since activation */
spinlock_t act_lock; /* protect updating of act_state */
u8 act_state; /* from XPC HB viewpoint */
+ u8 remote_vars_version; /* version# of partition's vars */
enum xpc_retval reason; /* reason partition is deactivating */
int reason_line; /* line# deactivation initiated from */
int reactivate_nasid; /* nasid in partition to reactivate */
+ unsigned long disengage_request_timeout; /* timeout in XPC_TICKS */
+ struct timer_list disengage_request_timer;
+
/* XPC infrastructure referencing and teardown control */
u8 nchannels; /* #of defined channels supported */
atomic_t nchannels_active; /* #of channels that are not DISCONNECTED */
+ atomic_t nchannels_engaged;/* #of channels engaged with remote part */
struct xpc_channel *channels;/* array of channel structures */
void *local_GPs_base; /* base address of kmalloc'd space */
#define XPC_P_TORNDOWN 0x03 /* infrastructure is torndown */
+
/*
* struct xpc_partition IPI_timer #of seconds to wait before checking for
* dropped IPIs. These occur whenever an IPI amo write doesn't complete until
#define XPC_P_DROPPED_IPI_WAIT (0.25 * HZ)
+/* number of seconds to wait for other partitions to disengage */
+#define XPC_DISENGAGE_REQUEST_TIMELIMIT 90
+
+/* interval in seconds to print 'waiting disengagement' messages */
+#define XPC_DISENGAGE_PRINTMSG_INTERVAL 10
+
+
#define XPC_PARTID(_p) ((partid_t) ((_p) - &xpc_partitions[0]))
/* found in xpc_partition.c */
extern int xpc_exiting;
-extern int xpc_hb_interval;
-extern int xpc_hb_check_interval;
extern struct xpc_vars *xpc_vars;
extern struct xpc_rsvd_page *xpc_rsvd_page;
extern struct xpc_vars_part *xpc_vars_part;
extern void xpc_allow_IPI_ops(void);
extern void xpc_restrict_IPI_ops(void);
extern int xpc_identify_act_IRQ_sender(void);
+extern int xpc_partition_disengaged(struct xpc_partition *);
extern enum xpc_retval xpc_mark_partition_active(struct xpc_partition *);
extern void xpc_mark_partition_inactive(struct xpc_partition *);
extern void xpc_discovery(void);
extern void xpc_deliver_msg(struct xpc_channel *);
extern void xpc_disconnect_channel(const int, struct xpc_channel *,
enum xpc_retval, unsigned long *);
-extern void xpc_disconnected_callout(struct xpc_channel *);
-extern void xpc_partition_down(struct xpc_partition *, enum xpc_retval);
+extern void xpc_disconnecting_callout(struct xpc_channel *);
+extern void xpc_partition_going_down(struct xpc_partition *, enum xpc_retval);
extern void xpc_teardown_infrastructure(struct xpc_partition *);
+/*
+ * This next set of inlines are used to keep track of when a partition is
+ * potentially engaged in accessing memory belonging to another partition.
+ */
+
+static inline void
+xpc_mark_partition_engaged(struct xpc_partition *part)
+{
+ unsigned long irq_flags;
+ AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+ (XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
+
+
+ local_irq_save(irq_flags);
+
+ /* set bit corresponding to our partid in remote partition's AMO */
+ FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
+ (1UL << sn_partition_id));
+ /*
+ * We must always use the nofault function regardless of whether we
+ * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+ * didn't, we'd never know that the other partition is down and would
+ * keep sending IPIs and AMOs to it until the heartbeat times out.
+ */
+ (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+ variable), xp_nofault_PIOR_target));
+
+ local_irq_restore(irq_flags);
+}
+
+static inline void
+xpc_mark_partition_disengaged(struct xpc_partition *part)
+{
+ unsigned long irq_flags;
+ AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+ (XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
+
+
+ local_irq_save(irq_flags);
+
+ /* clear bit corresponding to our partid in remote partition's AMO */
+ FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+ ~(1UL << sn_partition_id));
+ /*
+ * We must always use the nofault function regardless of whether we
+ * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+ * didn't, we'd never know that the other partition is down and would
+ * keep sending IPIs and AMOs to it until the heartbeat times out.
+ */
+ (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+ variable), xp_nofault_PIOR_target));
+
+ local_irq_restore(irq_flags);
+}
+
+static inline void
+xpc_request_partition_disengage(struct xpc_partition *part)
+{
+ unsigned long irq_flags;
+ AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+ (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
+
+
+ local_irq_save(irq_flags);
+
+ /* set bit corresponding to our partid in remote partition's AMO */
+ FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
+ (1UL << sn_partition_id));
+ /*
+ * We must always use the nofault function regardless of whether we
+ * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+ * didn't, we'd never know that the other partition is down and would
+ * keep sending IPIs and AMOs to it until the heartbeat times out.
+ */
+ (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+ variable), xp_nofault_PIOR_target));
+
+ local_irq_restore(irq_flags);
+}
+
+static inline void
+xpc_cancel_partition_disengage_request(struct xpc_partition *part)
+{
+ unsigned long irq_flags;
+ AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
+ (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
+
+
+ local_irq_save(irq_flags);
+
+ /* clear bit corresponding to our partid in remote partition's AMO */
+ FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+ ~(1UL << sn_partition_id));
+ /*
+ * We must always use the nofault function regardless of whether we
+ * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
+ * didn't, we'd never know that the other partition is down and would
+ * keep sending IPIs and AMOs to it until the heartbeat times out.
+ */
+ (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
+ variable), xp_nofault_PIOR_target));
+
+ local_irq_restore(irq_flags);
+}
+
+static inline u64
+xpc_partition_engaged(u64 partid_mask)
+{
+ AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
+
+
+ /* return our partition's AMO variable ANDed with partid_mask */
+ return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
+ partid_mask);
+}
+
+static inline u64
+xpc_partition_disengage_requested(u64 partid_mask)
+{
+ AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
+
+
+ /* return our partition's AMO variable ANDed with partid_mask */
+ return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
+ partid_mask);
+}
+
+static inline void
+xpc_clear_partition_engaged(u64 partid_mask)
+{
+ AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
+
+
+ /* clear bit(s) based on partid_mask in our partition's AMO */
+ FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+ ~partid_mask);
+}
+
+static inline void
+xpc_clear_partition_disengage_request(u64 partid_mask)
+{
+ AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
+
+
+ /* clear bit(s) based on partid_mask in our partition's AMO */
+ FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
+ ~partid_mask);
+}
+
+
+
/*
* The following set of macros and inlines are used for the sending and
* receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
* Flag the appropriate AMO variable and send an IPI to the specified node.
*/
static inline void
-xpc_activate_IRQ_send(u64 amos_page, int from_nasid, int to_nasid,
+xpc_activate_IRQ_send(u64 amos_page_pa, int from_nasid, int to_nasid,
int to_phys_cpuid)
{
int w_index = XPC_NASID_W_INDEX(from_nasid);
int b_index = XPC_NASID_B_INDEX(from_nasid);
- AMO_t *amos = (AMO_t *) __va(amos_page +
- (XP_MAX_PARTITIONS * sizeof(AMO_t)));
+ AMO_t *amos = (AMO_t *) __va(amos_page_pa +
+ (XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
(void) xpc_IPI_send(&amos[w_index], (1UL << b_index), to_nasid,
xpc_vars->act_nasid, xpc_vars->act_phys_cpuid);
}
+static inline void
+xpc_IPI_send_disengage(struct xpc_partition *part)
+{
+ xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
+ part->remote_act_nasid, part->remote_act_phys_cpuid);
+}
+
/*
* IPIs associated with SGI_XPC_NOTIFY IRQ.
* cacheable mapping for the entire region. This will prevent speculative
* reading of cached copies of our lines from being issued which will cause
* a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
- * (XP_MAX_PARTITIONS) AMO variables for message notification (xpc_main.c)
- * and an additional 16 AMO variables for partition activation (xpc_hb.c).
+ * (XP_MAX_PARTITIONS) AMO variables for message notification and an
+ * additional 16 (XP_NASID_MASK_WORDS) AMO variables for partition activation
+ * and 2 AMO variables for partition deactivation.
*/
static inline AMO_t *
-xpc_IPI_init(partid_t partid)
+xpc_IPI_init(int index)
{
- AMO_t *part_amo = xpc_vars->amos_page + partid;
+ AMO_t *amo = xpc_vars->amos_page + index;
- xpc_IPI_receive(part_amo);
- return part_amo;
+ (void) xpc_IPI_receive(amo); /* clear AMO variable */
+ return amo;
}
spin_lock_init(&ch->lock);
sema_init(&ch->msg_to_pull_sema, 1); /* mutex */
+ sema_init(&ch->wdisconnect_sema, 0); /* event wait */
atomic_set(&ch->n_on_msg_allocate_wq, 0);
init_waitqueue_head(&ch->msg_allocate_wq);
xpc_initialize_channels(part, partid);
atomic_set(&part->nchannels_active, 0);
+ atomic_set(&part->nchannels_engaged, 0);
/* local_IPI_amo were set to 0 by an earlier memset() */
sema_init(&ch->notify_queue[i].sema, 0);
}
- sema_init(&ch->teardown_sema, 0); /* event wait */
-
spin_lock_irqsave(&ch->lock, irq_flags);
ch->flags |= XPC_C_SETUP;
spin_unlock_irqrestore(&ch->lock, irq_flags);
}
+/*
+ * Notify those who wanted to be notified upon delivery of their message.
+ */
+static void
+xpc_notify_senders(struct xpc_channel *ch, enum xpc_retval reason, s64 put)
+{
+ struct xpc_notify *notify;
+ u8 notify_type;
+ s64 get = ch->w_remote_GP.get - 1;
+
+
+ while (++get < put && atomic_read(&ch->n_to_notify) > 0) {
+
+ notify = &ch->notify_queue[get % ch->local_nentries];
+
+ /*
+ * See if the notify entry indicates it was associated with
+ * a message who's sender wants to be notified. It is possible
+ * that it is, but someone else is doing or has done the
+ * notification.
+ */
+ notify_type = notify->type;
+ if (notify_type == 0 ||
+ cmpxchg(¬ify->type, notify_type, 0) !=
+ notify_type) {
+ continue;
+ }
+
+ DBUG_ON(notify_type != XPC_N_CALL);
+
+ 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",
+ (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);
+ }
+ }
+}
+
+
/*
* Free up message queues and other stuff that were allocated for the specified
* channel.
ch->remote_msgqueue = NULL;
kfree(ch->notify_queue);
ch->notify_queue = NULL;
-
- /* in case someone is waiting for the teardown to complete */
- up(&ch->teardown_sema);
}
}
xpc_process_disconnect(struct xpc_channel *ch, unsigned long *irq_flags)
{
struct xpc_partition *part = &xpc_partitions[ch->partid];
- u32 ch_flags = ch->flags;
+ u32 channel_was_connected = (ch->flags & XPC_C_WASCONNECTED);
DBUG_ON(!spin_is_locked(&ch->lock));
}
DBUG_ON(atomic_read(&ch->kthreads_assigned) != 0);
- /* it's now safe to free the channel's message queues */
-
- xpc_free_msgqueues(ch);
- DBUG_ON(ch->flags & XPC_C_SETUP);
+ if (part->act_state == XPC_P_DEACTIVATING) {
+ /* can't proceed until the other side disengages from us */
+ if (xpc_partition_engaged(1UL << ch->partid)) {
+ return;
+ }
- if (part->act_state != XPC_P_DEACTIVATING) {
+ } else {
/* as long as the other side is up do the full protocol */
}
}
+ /* wake those waiting for notify completion */
+ if (atomic_read(&ch->n_to_notify) > 0) {
+ /* >>> we do callout while holding ch->lock */
+ xpc_notify_senders(ch, ch->reason, ch->w_local_GP.put);
+ }
+
/* both sides are disconnected now */
- ch->flags = XPC_C_DISCONNECTED; /* clear all flags, but this one */
+ /* it's now safe to free the channel's message queues */
+ xpc_free_msgqueues(ch);
+
+ /* mark disconnected, clear all other flags except XPC_C_WDISCONNECT */
+ ch->flags = (XPC_C_DISCONNECTED | (ch->flags & XPC_C_WDISCONNECT));
atomic_dec(&part->nchannels_active);
- if (ch_flags & XPC_C_WASCONNECTED) {
+ if (channel_was_connected) {
dev_info(xpc_chan, "channel %d to partition %d disconnected, "
"reason=%d\n", ch->number, ch->partid, ch->reason);
}
+
+ /* wake the thread that is waiting for this channel to disconnect */
+ if (ch->flags & XPC_C_WDISCONNECT) {
+ spin_unlock_irqrestore(&ch->lock, *irq_flags);
+ up(&ch->wdisconnect_sema);
+ spin_lock_irqsave(&ch->lock, *irq_flags);
+ }
}
/*
* If RCLOSEREQUEST is set, we're probably waiting for
* RCLOSEREPLY. We should find it and a ROPENREQUEST packed
- * with this RCLOSEQREUQEST in the IPI_flags.
+ * with this RCLOSEREQUEST in the IPI_flags.
*/
if (ch->flags & XPC_C_RCLOSEREQUEST) {
"channel=%d\n", args->msg_size, args->local_nentries,
ch->partid, ch->number);
- if ((ch->flags & XPC_C_DISCONNECTING) ||
+ if ((ch->flags & (XPC_C_DISCONNECTING | XPC_C_WDISCONNECT)) ||
part->act_state == XPC_P_DEACTIVATING) {
spin_unlock_irqrestore(&ch->lock, irq_flags);
return;
}
-/*
- * Notify those who wanted to be notified upon delivery of their message.
- */
-static void
-xpc_notify_senders(struct xpc_channel *ch, enum xpc_retval reason, s64 put)
-{
- struct xpc_notify *notify;
- u8 notify_type;
- s64 get = ch->w_remote_GP.get - 1;
-
-
- while (++get < put && atomic_read(&ch->n_to_notify) > 0) {
-
- notify = &ch->notify_queue[get % ch->local_nentries];
-
- /*
- * See if the notify entry indicates it was associated with
- * a message who's sender wants to be notified. It is possible
- * that it is, but someone else is doing or has done the
- * notification.
- */
- notify_type = notify->type;
- if (notify_type == 0 ||
- cmpxchg(¬ify->type, notify_type, 0) !=
- notify_type) {
- continue;
- }
-
- DBUG_ON(notify_type != XPC_N_CALL);
-
- 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",
- (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);
- }
- }
-}
-
-
/*
* Clear some of the msg flags in the local message queue.
*/
u64 IPI_amo, IPI_flags;
struct xpc_channel *ch;
int ch_number;
+ u32 ch_flags;
IPI_amo = xpc_get_IPI_flags(part);
xpc_process_openclose_IPI(part, ch_number, IPI_flags);
}
+ ch_flags = ch->flags; /* need an atomic snapshot of flags */
- if (ch->flags & XPC_C_DISCONNECTING) {
+ if (ch_flags & XPC_C_DISCONNECTING) {
spin_lock_irqsave(&ch->lock, irq_flags);
xpc_process_disconnect(ch, &irq_flags);
spin_unlock_irqrestore(&ch->lock, irq_flags);
continue;
}
- if (!(ch->flags & XPC_C_CONNECTED)) {
- if (!(ch->flags & XPC_C_OPENREQUEST)) {
- DBUG_ON(ch->flags & XPC_C_SETUP);
+ if (!(ch_flags & XPC_C_CONNECTED)) {
+ if (!(ch_flags & XPC_C_OPENREQUEST)) {
+ DBUG_ON(ch_flags & XPC_C_SETUP);
(void) xpc_connect_channel(ch);
} else {
spin_lock_irqsave(&ch->lock, irq_flags);
/*
- * XPC's heartbeat code calls this function to inform XPC that a partition has
- * gone down. XPC responds by tearing down the XPartition Communication
+ * XPC's heartbeat code calls this function to inform XPC that a partition is
+ * going down. XPC responds by tearing down the XPartition Communication
* infrastructure used for the just downed partition.
*
* XPC's heartbeat code will never call this function and xpc_partition_up()
* at the same time.
*/
void
-xpc_partition_down(struct xpc_partition *part, enum xpc_retval reason)
+xpc_partition_going_down(struct xpc_partition *part, enum xpc_retval reason)
{
unsigned long irq_flags;
int ch_number;
}
- /* disconnect all channels associated with the downed partition */
+ /* disconnect channels associated with the partition going down */
for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
ch = &part->channels[ch_number];
-
xpc_msgqueue_ref(ch);
spin_lock_irqsave(&ch->lock, irq_flags);
* this partition.
*/
+ DBUG_ON(atomic_read(&part->nchannels_engaged) != 0);
DBUG_ON(atomic_read(&part->nchannels_active) != 0);
DBUG_ON(part->setup_state != XPC_P_SETUP);
part->setup_state = XPC_P_WTEARDOWN;
spin_lock_irqsave(&ch->lock, irq_flags);
- XPC_DISCONNECT_CHANNEL(ch, xpcUnregistering,
+ if (!(ch->flags & XPC_C_DISCONNECTED)) {
+ ch->flags |= XPC_C_WDISCONNECT;
+
+ XPC_DISCONNECT_CHANNEL(ch, xpcUnregistering,
&irq_flags);
+ }
spin_unlock_irqrestore(&ch->lock, irq_flags);
/*
* To disconnect a channel, and reflect it back to all who may be waiting.
*
- * >>> An OPEN is not allowed until XPC_C_DISCONNECTING is cleared by
- * >>> xpc_free_msgqueues().
+ * An OPEN is not allowed until XPC_C_DISCONNECTING is cleared by
+ * xpc_process_disconnect(), and if set, XPC_C_WDISCONNECT is cleared by
+ * xpc_disconnect_wait().
*
* THE CHANNEL IS TO BE LOCKED BY THE CALLER AND WILL REMAIN LOCKED UPON RETURN.
*/
xpc_disconnect_channel(const int line, struct xpc_channel *ch,
enum xpc_retval reason, unsigned long *irq_flags)
{
- u32 flags;
+ u32 channel_was_connected = (ch->flags & XPC_C_CONNECTED);
DBUG_ON(!spin_is_locked(&ch->lock));
XPC_SET_REASON(ch, reason, line);
- flags = ch->flags;
+ ch->flags |= (XPC_C_CLOSEREQUEST | XPC_C_DISCONNECTING);
/* some of these may not have been set */
ch->flags &= ~(XPC_C_OPENREQUEST | XPC_C_OPENREPLY |
XPC_C_ROPENREQUEST | XPC_C_ROPENREPLY |
XPC_C_CONNECTING | XPC_C_CONNECTED);
- ch->flags |= (XPC_C_CLOSEREQUEST | XPC_C_DISCONNECTING);
xpc_IPI_send_closerequest(ch, irq_flags);
- if (flags & XPC_C_CONNECTED) {
+ if (channel_was_connected) {
ch->flags |= XPC_C_WASCONNECTED;
}
+ spin_unlock_irqrestore(&ch->lock, *irq_flags);
+
+ /* wake all idle kthreads so they can exit */
if (atomic_read(&ch->kthreads_idle) > 0) {
- /* wake all idle kthreads so they can exit */
wake_up_all(&ch->idle_wq);
}
- spin_unlock_irqrestore(&ch->lock, *irq_flags);
-
-
/* wake those waiting to allocate an entry from the local msg queue */
-
if (atomic_read(&ch->n_on_msg_allocate_wq) > 0) {
wake_up(&ch->msg_allocate_wq);
}
- /* wake those waiting for notify completion */
-
- if (atomic_read(&ch->n_to_notify) > 0) {
- xpc_notify_senders(ch, reason, ch->w_local_GP.put);
- }
-
spin_lock_irqsave(&ch->lock, *irq_flags);
}
void
-xpc_disconnected_callout(struct xpc_channel *ch)
+xpc_disconnecting_callout(struct xpc_channel *ch)
{
/*
- * Let the channel's registerer know that the channel is now
+ * Let the channel's registerer know that the channel is being
* disconnected. We don't want to do this if the registerer was never
- * informed of a connection being made, unless the disconnect was for
- * abnormal reasons.
+ * informed of a connection being made.
*/
if (ch->func != NULL) {
- dev_dbg(xpc_chan, "ch->func() called, reason=%d, partid=%d, "
- "channel=%d\n", ch->reason, ch->partid, ch->number);
+ dev_dbg(xpc_chan, "ch->func() called, reason=xpcDisconnecting,"
+ " partid=%d, channel=%d\n", ch->partid, ch->number);
- ch->func(ch->reason, ch->partid, ch->number, NULL, ch->key);
+ ch->func(xpcDisconnecting, ch->partid, ch->number, NULL,
+ ch->key);
- dev_dbg(xpc_chan, "ch->func() returned, reason=%d, partid=%d, "
- "channel=%d\n", ch->reason, ch->partid, ch->number);
+ dev_dbg(xpc_chan, "ch->func() returned, reason="
+ "xpcDisconnecting, partid=%d, channel=%d\n",
+ ch->partid, ch->number);
}
}
xpc_notify_func func, void *key)
{
enum xpc_retval ret = xpcSuccess;
- struct xpc_notify *notify = NULL; // >>> to keep the compiler happy!!
+ struct xpc_notify *notify = notify;
s64 put, msg_number = msg->number;
#include <linux/interrupt.h>
#include <linux/slab.h>
#include <linux/delay.h>
+#include <linux/reboot.h>
#include <asm/sn/intr.h>
#include <asm/sn/sn_sal.h>
#include <asm/uaccess.h>
/* systune related variables for /proc/sys directories */
-static int xpc_hb_min = 1;
-static int xpc_hb_max = 10;
+static int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
+static int xpc_hb_min_interval = 1;
+static int xpc_hb_max_interval = 10;
-static int xpc_hb_check_min = 10;
-static int xpc_hb_check_max = 120;
+static int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_INTERVAL;
+static int xpc_hb_check_min_interval = 10;
+static int xpc_hb_check_max_interval = 120;
static ctl_table xpc_sys_xpc_hb_dir[] = {
{
&proc_dointvec_minmax,
&sysctl_intvec,
NULL,
- &xpc_hb_min, &xpc_hb_max
+ &xpc_hb_min_interval,
+ &xpc_hb_max_interval
},
{
2,
&proc_dointvec_minmax,
&sysctl_intvec,
NULL,
- &xpc_hb_check_min, &xpc_hb_check_max
+ &xpc_hb_check_min_interval,
+ &xpc_hb_check_max_interval
},
{0}
};
static unsigned long xpc_hb_check_timeout;
-/* xpc_hb_checker thread exited notification */
-static DECLARE_MUTEX_LOCKED(xpc_hb_checker_exited);
+/* used as an indication of when the xpc_hb_checker thread is inactive */
+static DECLARE_MUTEX_LOCKED(xpc_hb_checker_inactive);
-/* xpc_discovery thread exited notification */
-static DECLARE_MUTEX_LOCKED(xpc_discovery_exited);
+/* used as an indication of when the xpc_discovery thread is inactive */
+static DECLARE_MUTEX_LOCKED(xpc_discovery_inactive);
static struct timer_list xpc_hb_timer;
static void xpc_kthread_waitmsgs(struct xpc_partition *, struct xpc_channel *);
+static int xpc_system_reboot(struct notifier_block *, unsigned long, void *);
+static struct notifier_block xpc_reboot_notifier = {
+ .notifier_call = xpc_system_reboot,
+};
+
+
+/*
+ * Timer function to enforce the timelimit on the partition disengage request.
+ */
+static void
+xpc_timeout_partition_disengage_request(unsigned long data)
+{
+ struct xpc_partition *part = (struct xpc_partition *) data;
+
+
+ DBUG_ON(XPC_TICKS < part->disengage_request_timeout);
+
+ (void) xpc_partition_disengaged(part);
+
+ DBUG_ON(part->disengage_request_timeout != 0);
+ DBUG_ON(xpc_partition_engaged(1UL << XPC_PARTID(part)) != 0);
+}
+
+
/*
* Notify the heartbeat check thread that an IRQ has been received.
*/
while (!(volatile int) xpc_exiting) {
- /* wait for IRQ or timeout */
- (void) wait_event_interruptible(xpc_act_IRQ_wq,
- (last_IRQ_count < atomic_read(&xpc_act_IRQ_rcvd) ||
- jiffies >= xpc_hb_check_timeout ||
- (volatile int) xpc_exiting));
-
dev_dbg(xpc_part, "woke up with %d ticks rem; %d IRQs have "
"been received\n",
(int) (xpc_hb_check_timeout - jiffies),
}
+ /* check for outstanding IRQs */
new_IRQ_count = atomic_read(&xpc_act_IRQ_rcvd);
if (last_IRQ_count < new_IRQ_count || force_IRQ != 0) {
force_IRQ = 0;
xpc_hb_check_timeout = jiffies +
(xpc_hb_check_interval * HZ);
}
+
+ /* wait for IRQ or timeout */
+ (void) wait_event_interruptible(xpc_act_IRQ_wq,
+ (last_IRQ_count < atomic_read(&xpc_act_IRQ_rcvd) ||
+ jiffies >= xpc_hb_check_timeout ||
+ (volatile int) xpc_exiting));
}
dev_dbg(xpc_part, "heartbeat checker is exiting\n");
/* mark this thread as inactive */
- up(&xpc_hb_checker_exited);
+ up(&xpc_hb_checker_inactive);
return 0;
}
dev_dbg(xpc_part, "discovery thread is exiting\n");
/* mark this thread as inactive */
- up(&xpc_discovery_exited);
+ up(&xpc_discovery_inactive);
return 0;
}
"partition %d\n", XPC_PARTID(part));
/* wait a 1/4 of a second or so */
- msleep_interruptible(250);
+ (void) msleep_interruptible(250);
if (part->act_state == XPC_P_DEACTIVATING) {
return part->reason;
xpc_channel_mgr(struct xpc_partition *part)
{
while (part->act_state != XPC_P_DEACTIVATING ||
- atomic_read(&part->nchannels_active) > 0) {
+ atomic_read(&part->nchannels_active) > 0 ||
+ !xpc_partition_disengaged(part)) {
xpc_process_channel_activity(part);
(volatile u64) part->local_IPI_amo != 0 ||
((volatile u8) part->act_state ==
XPC_P_DEACTIVATING &&
- atomic_read(&part->nchannels_active) == 0)));
+ atomic_read(&part->nchannels_active) == 0 &&
+ xpc_partition_disengaged(part))));
atomic_set(&part->channel_mgr_requests, 1);
// >>> Does it need to wakeup periodically as well? In case we
return 0;
}
- XPC_ALLOW_HB(partid, xpc_vars);
+ xpc_allow_hb(partid, xpc_vars);
xpc_IPI_send_activated(part);
*/
(void) xpc_partition_up(part);
+ xpc_disallow_hb(partid, xpc_vars);
xpc_mark_partition_inactive(part);
if (part->reason == xpcReactivating) {
xpc_kthread_waitmsgs(part, ch);
}
- if (atomic_dec_return(&ch->kthreads_assigned) == 0 &&
- ((ch->flags & XPC_C_CONNECTCALLOUT) ||
- (ch->reason != xpcUnregistering &&
- ch->reason != xpcOtherUnregistering))) {
- xpc_disconnected_callout(ch);
+ if (atomic_dec_return(&ch->kthreads_assigned) == 0) {
+ if (ch->flags & XPC_C_CONNECTCALLOUT) {
+ xpc_disconnecting_callout(ch);
+ }
+ if (atomic_dec_return(&part->nchannels_engaged) == 0) {
+ xpc_mark_partition_disengaged(part);
+ xpc_IPI_send_disengage(part);
+ }
}
unsigned long irq_flags;
pid_t pid;
u64 args = XPC_PACK_ARGS(ch->partid, ch->number);
+ struct xpc_partition *part = &xpc_partitions[ch->partid];
while (needed-- > 0) {
* kthread. That kthread is responsible for doing the
* counterpart to the following before it exits.
*/
- (void) xpc_part_ref(&xpc_partitions[ch->partid]);
+ (void) xpc_part_ref(part);
xpc_msgqueue_ref(ch);
- atomic_inc(&ch->kthreads_assigned);
+ if (atomic_inc_return(&ch->kthreads_assigned) == 1) {
+ if (atomic_inc_return(&part->nchannels_engaged) == 1) {
+ xpc_mark_partition_engaged(part);
+ }
+ }
ch->kthreads_created++; // >>> temporary debug only!!!
}
}
void
xpc_disconnect_wait(int ch_number)
{
+ unsigned long irq_flags;
partid_t partid;
struct xpc_partition *part;
struct xpc_channel *ch;
if (xpc_part_ref(part)) {
ch = &part->channels[ch_number];
-// >>> how do we keep from falling into the window between our check and going
-// >>> down and coming back up where sema is re-inited?
- if (ch->flags & XPC_C_SETUP) {
- (void) down(&ch->teardown_sema);
+ if (ch->flags & XPC_C_WDISCONNECT) {
+ if (!(ch->flags & XPC_C_DISCONNECTED)) {
+ (void) down(&ch->wdisconnect_sema);
+ }
+ spin_lock_irqsave(&ch->lock, irq_flags);
+ ch->flags &= ~XPC_C_WDISCONNECT;
+ spin_unlock_irqrestore(&ch->lock, irq_flags);
}
xpc_part_deref(part);
static void
-xpc_do_exit(void)
+xpc_do_exit(enum xpc_retval reason)
{
partid_t partid;
int active_part_count;
struct xpc_partition *part;
+ unsigned long printmsg_time;
- /* now it's time to eliminate our heartbeat */
- del_timer_sync(&xpc_hb_timer);
- xpc_vars->heartbeating_to_mask = 0;
-
- /* indicate to others that our reserved page is uninitialized */
- xpc_rsvd_page->vars_pa = 0;
+ /* a 'rmmod XPC' and a 'reboot' cannot both end up here together */
+ DBUG_ON(xpc_exiting == 1);
/*
- * Ignore all incoming interrupts. Without interupts the heartbeat
- * checker won't activate any new partitions that may come up.
- */
- free_irq(SGI_XPC_ACTIVATE, NULL);
-
- /*
- * Cause the heartbeat checker and the discovery threads to exit.
- * We don't want them attempting to activate new partitions as we
- * try to deactivate the existing ones.
+ * Let the heartbeat checker thread and the discovery thread
+ * (if one is running) know that they should exit. Also wake up
+ * the heartbeat checker thread in case it's sleeping.
*/
xpc_exiting = 1;
wake_up_interruptible(&xpc_act_IRQ_wq);
- /* wait for the heartbeat checker thread to mark itself inactive */
- down(&xpc_hb_checker_exited);
+ /* ignore all incoming interrupts */
+ free_irq(SGI_XPC_ACTIVATE, NULL);
/* wait for the discovery thread to mark itself inactive */
- down(&xpc_discovery_exited);
+ down(&xpc_discovery_inactive);
+
+ /* wait for the heartbeat checker thread to mark itself inactive */
+ down(&xpc_hb_checker_inactive);
- msleep_interruptible(300);
+ /* sleep for a 1/3 of a second or so */
+ (void) msleep_interruptible(300);
/* wait for all partitions to become inactive */
+ printmsg_time = jiffies;
+
do {
active_part_count = 0;
for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
part = &xpc_partitions[partid];
- if (part->act_state != XPC_P_INACTIVE) {
- active_part_count++;
-
- XPC_DEACTIVATE_PARTITION(part, xpcUnloading);
+ if (xpc_partition_disengaged(part) &&
+ part->act_state == XPC_P_INACTIVE) {
+ continue;
}
+
+ active_part_count++;
+
+ XPC_DEACTIVATE_PARTITION(part, reason);
+ }
+
+ if (active_part_count == 0) {
+ break;
+ }
+
+ if (jiffies >= printmsg_time) {
+ dev_info(xpc_part, "waiting for partitions to "
+ "deactivate/disengage, active count=%d, remote "
+ "engaged=0x%lx\n", active_part_count,
+ xpc_partition_engaged(1UL << partid));
+
+ printmsg_time = jiffies +
+ (XPC_DISENGAGE_PRINTMSG_INTERVAL * HZ);
}
- if (active_part_count)
- msleep_interruptible(300);
- } while (active_part_count > 0);
+ /* sleep for a 1/3 of a second or so */
+ (void) msleep_interruptible(300);
+
+ } while (1);
+
+ DBUG_ON(xpc_partition_engaged(-1UL));
+
+
+ /* indicate to others that our reserved page is uninitialized */
+ xpc_rsvd_page->vars_pa = 0;
+
+ /* now it's time to eliminate our heartbeat */
+ del_timer_sync(&xpc_hb_timer);
+ DBUG_ON(xpc_vars->heartbeating_to_mask == 0);
+ /* take ourselves off of the reboot_notifier_list */
+ (void) unregister_reboot_notifier(&xpc_reboot_notifier);
/* close down protections for IPI operations */
xpc_restrict_IPI_ops();
}
+/*
+ * This function is called when the system is being rebooted.
+ */
+static int
+xpc_system_reboot(struct notifier_block *nb, unsigned long event, void *unused)
+{
+ enum xpc_retval reason;
+
+
+ switch (event) {
+ case SYS_RESTART:
+ reason = xpcSystemReboot;
+ break;
+ case SYS_HALT:
+ reason = xpcSystemHalt;
+ break;
+ case SYS_POWER_OFF:
+ reason = xpcSystemPoweroff;
+ break;
+ default:
+ reason = xpcSystemGoingDown;
+ }
+
+ xpc_do_exit(reason);
+ return NOTIFY_DONE;
+}
+
+
int __init
xpc_init(void)
{
spin_lock_init(&part->act_lock);
part->act_state = XPC_P_INACTIVE;
XPC_SET_REASON(part, 0, 0);
+
+ init_timer(&part->disengage_request_timer);
+ part->disengage_request_timer.function =
+ xpc_timeout_partition_disengage_request;
+ part->disengage_request_timer.data = (unsigned long) part;
+
part->setup_state = XPC_P_UNSET;
init_waitqueue_head(&part->teardown_wq);
atomic_set(&part->references, 0);
}
+ /* add ourselves to the reboot_notifier_list */
+ ret = register_reboot_notifier(&xpc_reboot_notifier);
+ if (ret != 0) {
+ dev_warn(xpc_part, "can't register reboot notifier\n");
+ }
+
+
/*
* Set the beating to other partitions into motion. This is
* the last requirement for other partitions' discovery to
/* indicate to others that our reserved page is uninitialized */
xpc_rsvd_page->vars_pa = 0;
+ /* take ourselves off of the reboot_notifier_list */
+ (void) unregister_reboot_notifier(&xpc_reboot_notifier);
+
del_timer_sync(&xpc_hb_timer);
free_irq(SGI_XPC_ACTIVATE, NULL);
xpc_restrict_IPI_ops();
dev_err(xpc_part, "failed while forking discovery thread\n");
/* mark this new thread as a non-starter */
- up(&xpc_discovery_exited);
+ up(&xpc_discovery_inactive);
- xpc_do_exit();
+ xpc_do_exit(xpcUnloading);
return -EBUSY;
}
void __exit
xpc_exit(void)
{
- xpc_do_exit();
+ xpc_do_exit(xpcUnloading);
}
module_exit(xpc_exit);
xpc_remote_copy_buffer[XPC_RSVD_PAGE_ALIGNED_SIZE];
-/* systune related variables */
-int xpc_hb_interval = XPC_HB_DEFAULT_INTERVAL;
-int xpc_hb_check_interval = XPC_HB_CHECK_DEFAULT_TIMEOUT;
-
-
/*
* Given a nasid, get the physical address of the partition's reserved page
* for that nasid. This function returns 0 on any error.
xpc_vars->amos_page = amos_page; /* save for next load of XPC */
- /*
- * Initialize the activation related AMO variables.
- */
- xpc_vars->act_amos = xpc_IPI_init(XP_MAX_PARTITIONS);
- for (i = 1; i < XP_NASID_MASK_WORDS; i++) {
- xpc_IPI_init(i + XP_MAX_PARTITIONS);
+ /* initialize the activate IRQ related AMO variables */
+ for (i = 0; i < XP_NASID_MASK_WORDS; i++) {
+ (void) xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
}
+
+ /* initialize the engaged remote partitions related AMO variables */
+ (void) xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
+ (void) xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);
+
/* export AMO page's physical address to other partitions */
xpc_vars->amos_page_pa = ia64_tpa((u64) xpc_vars->amos_page);
+ /* timestamp of when reserved page was initialized */
+ rp->stamp = CURRENT_TIME;
+
/*
* This signifies to the remote partition that our reserved
* page is initialized.
remote_vars = (struct xpc_vars *) xpc_remote_copy_buffer;
for (partid = 1; partid < XP_MAX_PARTITIONS; partid++) {
+
+ if (xpc_exiting) {
+ break;
+ }
+
if (partid == sn_partition_id) {
continue;
}
if (((remote_vars->heartbeat == part->last_heartbeat) &&
(remote_vars->kdb_status == 0)) ||
- !XPC_HB_ALLOWED(sn_partition_id, remote_vars)) {
+ !xpc_hb_allowed(sn_partition_id, remote_vars)) {
XPC_DEACTIVATE_PARTITION(part, xpcNoHeartbeat);
continue;
*/
static enum xpc_retval
xpc_get_remote_rp(int nasid, u64 *discovered_nasids,
- struct xpc_rsvd_page *remote_rp, u64 *remote_rsvd_page_pa)
+ struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
{
int bres, i;
/* get the reserved page's physical address */
- *remote_rsvd_page_pa = xpc_get_rsvd_page_pa(nasid, (u64) remote_rp,
+ *remote_rp_pa = xpc_get_rsvd_page_pa(nasid, (u64) remote_rp,
XPC_RSVD_PAGE_ALIGNED_SIZE);
- if (*remote_rsvd_page_pa == 0) {
+ if (*remote_rp_pa == 0) {
return xpcNoRsvdPageAddr;
}
/* pull over the reserved page structure */
- bres = xp_bte_copy(*remote_rsvd_page_pa, ia64_tpa((u64) remote_rp),
+ bres = xp_bte_copy(*remote_rp_pa, ia64_tpa((u64) remote_rp),
XPC_RSVD_PAGE_ALIGNED_SIZE,
(BTE_NOTIFY | BTE_WACQUIRE), NULL);
if (bres != BTE_SUCCESS) {
}
+/*
+ * Update the remote partition's info.
+ */
+static void
+xpc_update_partition_info(struct xpc_partition *part, u8 remote_rp_version,
+ struct timespec *remote_rp_stamp, u64 remote_rp_pa,
+ u64 remote_vars_pa, struct xpc_vars *remote_vars)
+{
+ part->remote_rp_version = remote_rp_version;
+ dev_dbg(xpc_part, " remote_rp_version = 0x%016lx\n",
+ part->remote_rp_version);
+
+ part->remote_rp_stamp = *remote_rp_stamp;
+ dev_dbg(xpc_part, " remote_rp_stamp (tv_sec = 0x%lx tv_nsec = 0x%lx\n",
+ part->remote_rp_stamp.tv_sec, part->remote_rp_stamp.tv_nsec);
+
+ part->remote_rp_pa = remote_rp_pa;
+ dev_dbg(xpc_part, " remote_rp_pa = 0x%016lx\n", part->remote_rp_pa);
+
+ part->remote_vars_pa = remote_vars_pa;
+ dev_dbg(xpc_part, " remote_vars_pa = 0x%016lx\n",
+ part->remote_vars_pa);
+
+ part->last_heartbeat = remote_vars->heartbeat;
+ dev_dbg(xpc_part, " last_heartbeat = 0x%016lx\n",
+ part->last_heartbeat);
+
+ part->remote_vars_part_pa = remote_vars->vars_part_pa;
+ dev_dbg(xpc_part, " remote_vars_part_pa = 0x%016lx\n",
+ part->remote_vars_part_pa);
+
+ part->remote_act_nasid = remote_vars->act_nasid;
+ dev_dbg(xpc_part, " remote_act_nasid = 0x%x\n",
+ part->remote_act_nasid);
+
+ part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
+ dev_dbg(xpc_part, " remote_act_phys_cpuid = 0x%x\n",
+ part->remote_act_phys_cpuid);
+
+ part->remote_amos_page_pa = remote_vars->amos_page_pa;
+ dev_dbg(xpc_part, " remote_amos_page_pa = 0x%lx\n",
+ part->remote_amos_page_pa);
+
+ part->remote_vars_version = remote_vars->version;
+ dev_dbg(xpc_part, " remote_vars_version = 0x%x\n",
+ part->remote_vars_version);
+}
+
+
/*
* Prior code has determine the nasid which generated an IPI. Inspect
* that nasid to determine if its partition needs to be activated or
{
struct xpc_rsvd_page *remote_rp;
struct xpc_vars *remote_vars;
- u64 remote_rsvd_page_pa;
+ u64 remote_rp_pa;
u64 remote_vars_pa;
+ int remote_rp_version;
+ int reactivate = 0;
+ int stamp_diff;
+ struct timespec remote_rp_stamp = { 0, 0 };
partid_t partid;
struct xpc_partition *part;
enum xpc_retval ret;
remote_rp = (struct xpc_rsvd_page *) xpc_remote_copy_buffer;
- ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rsvd_page_pa);
+ ret = xpc_get_remote_rp(nasid, NULL, remote_rp, &remote_rp_pa);
if (ret != xpcSuccess) {
dev_warn(xpc_part, "unable to get reserved page from nasid %d, "
"which sent interrupt, reason=%d\n", nasid, ret);
}
remote_vars_pa = remote_rp->vars_pa;
+ remote_rp_version = remote_rp->version;
+ if (XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
+ remote_rp_stamp = remote_rp->stamp;
+ }
partid = remote_rp->partid;
part = &xpc_partitions[partid];
"%ld:0x%lx\n", (int) nasid, (int) partid, part->act_IRQ_rcvd,
remote_vars->heartbeat, remote_vars->heartbeating_to_mask);
+ if (xpc_partition_disengaged(part) &&
+ part->act_state == XPC_P_INACTIVE) {
- if (part->act_state == XPC_P_INACTIVE) {
+ xpc_update_partition_info(part, remote_rp_version,
+ &remote_rp_stamp, remote_rp_pa,
+ remote_vars_pa, remote_vars);
- part->remote_rp_pa = remote_rsvd_page_pa;
- dev_dbg(xpc_part, " remote_rp_pa = 0x%016lx\n",
- part->remote_rp_pa);
+ if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
+ if (xpc_partition_disengage_requested(1UL << partid)) {
+ /*
+ * Other side is waiting on us to disengage,
+ * even though we already have.
+ */
+ return;
+ }
+ } else {
+ /* other side doesn't support disengage requests */
+ xpc_clear_partition_disengage_request(1UL << partid);
+ }
- part->remote_vars_pa = remote_vars_pa;
- dev_dbg(xpc_part, " remote_vars_pa = 0x%016lx\n",
- part->remote_vars_pa);
+ xpc_activate_partition(part);
+ return;
+ }
- part->last_heartbeat = remote_vars->heartbeat;
- dev_dbg(xpc_part, " last_heartbeat = 0x%016lx\n",
- part->last_heartbeat);
+ DBUG_ON(part->remote_rp_version == 0);
+ DBUG_ON(part->remote_vars_version == 0);
+
+ if (!XPC_SUPPORTS_RP_STAMP(part->remote_rp_version)) {
+ DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(part->
+ remote_vars_version));
+
+ if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
+ DBUG_ON(XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
+ version));
+ /* see if the other side rebooted */
+ if (part->remote_amos_page_pa ==
+ remote_vars->amos_page_pa &&
+ xpc_hb_allowed(sn_partition_id,
+ remote_vars)) {
+ /* doesn't look that way, so ignore the IPI */
+ return;
+ }
+ }
- part->remote_vars_part_pa = remote_vars->vars_part_pa;
- dev_dbg(xpc_part, " remote_vars_part_pa = 0x%016lx\n",
- part->remote_vars_part_pa);
+ /*
+ * Other side rebooted and previous XPC didn't support the
+ * disengage request, so we don't need to do anything special.
+ */
- part->remote_act_nasid = remote_vars->act_nasid;
- dev_dbg(xpc_part, " remote_act_nasid = 0x%x\n",
- part->remote_act_nasid);
+ xpc_update_partition_info(part, remote_rp_version,
+ &remote_rp_stamp, remote_rp_pa,
+ remote_vars_pa, remote_vars);
+ part->reactivate_nasid = nasid;
+ XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
+ return;
+ }
- part->remote_act_phys_cpuid = remote_vars->act_phys_cpuid;
- dev_dbg(xpc_part, " remote_act_phys_cpuid = 0x%x\n",
- part->remote_act_phys_cpuid);
+ DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version));
- part->remote_amos_page_pa = remote_vars->amos_page_pa;
- dev_dbg(xpc_part, " remote_amos_page_pa = 0x%lx\n",
- part->remote_amos_page_pa);
+ if (!XPC_SUPPORTS_RP_STAMP(remote_rp_version)) {
+ DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
- xpc_activate_partition(part);
+ /*
+ * Other side rebooted and previous XPC did support the
+ * disengage request, but the new one doesn't.
+ */
- } else if (part->remote_amos_page_pa != remote_vars->amos_page_pa ||
- !XPC_HB_ALLOWED(sn_partition_id, remote_vars)) {
+ xpc_clear_partition_engaged(1UL << partid);
+ xpc_clear_partition_disengage_request(1UL << partid);
+ xpc_update_partition_info(part, remote_rp_version,
+ &remote_rp_stamp, remote_rp_pa,
+ remote_vars_pa, remote_vars);
+ reactivate = 1;
+
+ } else {
+ DBUG_ON(!XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->version));
+
+ stamp_diff = xpc_compare_stamps(&part->remote_rp_stamp,
+ &remote_rp_stamp);
+ if (stamp_diff != 0) {
+ DBUG_ON(stamp_diff >= 0);
+
+ /*
+ * Other side rebooted and the previous XPC did support
+ * the disengage request, as does the new one.
+ */
+
+ DBUG_ON(xpc_partition_engaged(1UL << partid));
+ DBUG_ON(xpc_partition_disengage_requested(1UL <<
+ partid));
+
+ xpc_update_partition_info(part, remote_rp_version,
+ &remote_rp_stamp, remote_rp_pa,
+ remote_vars_pa, remote_vars);
+ reactivate = 1;
+ }
+ }
+
+ if (!xpc_partition_disengaged(part)) {
+ /* still waiting on other side to disengage from us */
+ return;
+ }
+
+ if (reactivate) {
part->reactivate_nasid = nasid;
XPC_DEACTIVATE_PARTITION(part, xpcReactivating);
+
+ } else if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version) &&
+ xpc_partition_disengage_requested(1UL << partid)) {
+ XPC_DEACTIVATE_PARTITION(part, xpcOtherGoingDown);
}
}
struct xpc_rsvd_page *rp = (struct xpc_rsvd_page *) xpc_rsvd_page;
- act_amos = xpc_vars->act_amos;
+ act_amos = xpc_vars->amos_page + XPC_ACTIVATE_IRQ_AMOS;
/* scan through act AMO variable looking for non-zero entries */
for (word = 0; word < XP_NASID_MASK_WORDS; word++) {
+ if (xpc_exiting) {
+ break;
+ }
+
nasid_mask = xpc_IPI_receive(&act_amos[word]);
if (nasid_mask == 0) {
/* no IRQs from nasids in this variable */
}
+/*
+ * See if the other side has responded to a partition disengage request
+ * from us.
+ */
+int
+xpc_partition_disengaged(struct xpc_partition *part)
+{
+ partid_t partid = XPC_PARTID(part);
+ int disengaged;
+
+
+ disengaged = (xpc_partition_engaged(1UL << partid) == 0);
+ if (part->disengage_request_timeout) {
+ if (!disengaged) {
+ if (jiffies < part->disengage_request_timeout) {
+ /* timelimit hasn't been reached yet */
+ return 0;
+ }
+
+ /*
+ * Other side hasn't responded to our disengage
+ * request in a timely fashion, so assume it's dead.
+ */
+
+ xpc_clear_partition_engaged(1UL << partid);
+ disengaged = 1;
+ }
+ part->disengage_request_timeout = 0;
+
+ /* cancel the timer function, provided it's not us */
+ if (!in_interrupt()) {
+ del_singleshot_timer_sync(&part->
+ disengage_request_timer);
+ }
+
+ DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
+ part->act_state != XPC_P_INACTIVE);
+ if (part->act_state != XPC_P_INACTIVE) {
+ xpc_wakeup_channel_mgr(part);
+ }
+
+ if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
+ xpc_cancel_partition_disengage_request(part);
+ }
+ }
+ return disengaged;
+}
+
+
/*
* Mark specified partition as active.
*/
enum xpc_retval reason)
{
unsigned long irq_flags;
- partid_t partid = XPC_PARTID(part);
spin_lock_irqsave(&part->act_lock, irq_flags);
spin_unlock_irqrestore(&part->act_lock, irq_flags);
- XPC_DISALLOW_HB(partid, xpc_vars);
+ if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
+ xpc_request_partition_disengage(part);
+ xpc_IPI_send_disengage(part);
+
+ /* set a timelimit on the disengage request */
+ part->disengage_request_timeout = jiffies +
+ (XPC_DISENGAGE_REQUEST_TIMELIMIT * HZ);
+ part->disengage_request_timer.expires =
+ part->disengage_request_timeout;
+ add_timer(&part->disengage_request_timer);
+ }
dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n", partid,
reason);
- xpc_partition_down(part, reason);
+ xpc_partition_going_down(part, reason);
}
/*
- * Mark specified partition as active.
+ * Mark specified partition as inactive.
*/
void
xpc_mark_partition_inactive(struct xpc_partition *part)
void *remote_rp_base;
struct xpc_rsvd_page *remote_rp;
struct xpc_vars *remote_vars;
- u64 remote_rsvd_page_pa;
+ u64 remote_rp_pa;
u64 remote_vars_pa;
int region;
int max_regions;
/* pull over the reserved page structure */
ret = xpc_get_remote_rp(nasid, discovered_nasids,
- remote_rp, &remote_rsvd_page_pa);
+ remote_rp, &remote_rp_pa);
if (ret != xpcSuccess) {
dev_dbg(xpc_part, "unable to get reserved page "
"from nasid %d, reason=%d\n", nasid,
remote_vars->act_nasid,
remote_vars->act_phys_cpuid);
+ if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
+ version)) {
+ part->remote_amos_page_pa =
+ remote_vars->amos_page_pa;
+ xpc_mark_partition_disengaged(part);
+ xpc_cancel_partition_disengage_request(part);
+ }
xpc_IPI_send_activate(remote_vars);
}
}
xpcInvalidPartid, /* 42: invalid partition ID */
xpcLocalPartid, /* 43: local partition ID */
- xpcUnknownReason /* 44: unknown reason -- must be last in list */
+ xpcOtherGoingDown, /* 44: other side going down, reason unknown */
+ xpcSystemGoingDown, /* 45: system is going down, reason unknown */
+ xpcSystemHalt, /* 46: system is being halted */
+ xpcSystemReboot, /* 47: system is being rebooted */
+ xpcSystemPoweroff, /* 48: system is being powered off */
+
+ xpcDisconnecting, /* 49: channel disconnecting (closing) */
+
+ xpcUnknownReason /* 50: unknown reason -- must be last in list */
};