Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit a607c389 authored by Dean Nelson's avatar Dean Nelson Committed by Tony Luck
Browse files

[IA64-SGI] get XPC to cleanly disengage from remote memory references 



When XPC is being shutdown (i.e., rmmod, reboot) it doesn't ensure that
other partitions with whom it was connected have completely disengaged
from any attempt at cross-partition memory references. This can lead to
MCAs in any of these other partitions when the partition is reset.

Signed-off-by: default avatarDean Nelson <dcn@sgi.com>
Signed-off-by: default avatarTony Luck <tony.luck@intel.com>
parent 4706df3d

arch/ia64/sn/kernel/xpc.h

+265 −23
Original line number Diff line number Diff line
@@ -57,7 +57,7 @@ 
#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"
@@ -67,11 +67,6 @@ 
#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.

 *
@@ -88,14 +83,38 @@ struct xpc_rsvd_page { 
	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
@@ -121,12 +140,61 @@ struct xpc_vars { 
	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.

 *
@@ -358,7 +426,7 @@ struct xpc_channel { 
	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 */
@@ -410,6 +478,7 @@ struct xpc_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 */
@@ -422,6 +491,8 @@ struct xpc_partition { 


	/* 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 */
@@ -432,10 +503,14 @@ struct xpc_partition { 
	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 */
@@ -454,6 +529,7 @@ struct xpc_partition { 


	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 */
@@ -518,6 +594,7 @@ struct xpc_partition { 
#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
@@ -526,6 +603,13 @@ struct xpc_partition { 
#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]))
@@ -550,8 +634,6 @@ extern void xpc_activate_partition(struct xpc_partition *); 


/* 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;
@@ -561,6 +643,7 @@ extern struct xpc_rsvd_page *xpc_rsvd_page_init(void); 
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);
@@ -585,8 +668,8 @@ extern void xpc_connected_callout(struct xpc_channel *); 
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 *);
@@ -673,6 +756,157 @@ xpc_part_ref(struct xpc_partition *part) 






/*

 * 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,
@@ -722,13 +956,13 @@ xpc_IPI_send(AMO_t *amo, u64 flag, int nasid, int phys_cpuid, int vector) 
 * 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,
@@ -756,6 +990,13 @@ xpc_IPI_send_reactivate(struct xpc_partition *part) 
				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.
@@ -903,17 +1144,18 @@ xpc_IPI_send_local_msgrequest(struct xpc_channel *ch) 
 * 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;

}

arch/ia64/sn/kernel/xpc_channel.c

+115 −101

File changed.

Preview size limit exceeded, changes collapsed.

arch/ia64/sn/kernel/xpc_main.c

+179 −63
Original line number Diff line number Diff line
@@ -54,6 +54,7 @@ 
#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>
@@ -82,11 +83,13 @@ struct device *xpc_chan = &xpc_chan_dbg_subname; 


/* 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[] = {

	{
@@ -99,7 +102,8 @@ 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,
@@ -111,7 +115,8 @@ static ctl_table xpc_sys_xpc_hb_dir[] = { 
		&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}

};
@@ -148,11 +153,11 @@ static DECLARE_WAIT_QUEUE_HEAD(xpc_act_IRQ_wq); 


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;
@@ -161,6 +166,30 @@ 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.

 */
@@ -214,12 +243,6 @@ xpc_hb_checker(void *ignore) 


	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),
@@ -240,6 +263,7 @@ xpc_hb_checker(void *ignore) 
		}





		/* 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;
@@ -257,13 +281,19 @@ xpc_hb_checker(void *ignore) 
			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;

}
@@ -283,7 +313,7 @@ xpc_initiate_discovery(void *ignore) 
	dev_dbg(xpc_part, "discovery thread is exiting\n");



	/* mark this thread as inactive */

	up(&xpc_discovery_exited);

	up(&xpc_discovery_inactive);

	return 0;

}
@@ -309,7 +339,7 @@ xpc_make_first_contact(struct xpc_partition *part) 
			"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;
@@ -336,7 +366,8 @@ static void 
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);
@@ -360,7 +391,8 @@ xpc_channel_mgr(struct xpc_partition *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
@@ -482,7 +514,7 @@ xpc_activating(void *__partid) 
		return 0;

	}



	XPC_ALLOW_HB(partid, xpc_vars);

	xpc_allow_hb(partid, xpc_vars);

	xpc_IPI_send_activated(part);
@@ -492,6 +524,7 @@ xpc_activating(void *__partid) 
	 */

	(void) xpc_partition_up(part);



	xpc_disallow_hb(partid, xpc_vars);

	xpc_mark_partition_inactive(part);



	if (part->reason == xpcReactivating) {
@@ -704,11 +737,14 @@ xpc_daemonize_kthread(void *args) 
		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);

		}

	}
@@ -740,6 +776,7 @@ xpc_create_kthreads(struct xpc_channel *ch, int needed) 
	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) {
@@ -770,9 +807,13 @@ xpc_create_kthreads(struct xpc_channel *ch, int needed) 
		 * 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!!!

	}

}
@@ -781,6 +822,7 @@ xpc_create_kthreads(struct xpc_channel *ch, int needed) 
void

xpc_disconnect_wait(int ch_number)

{

	unsigned long irq_flags;

	partid_t partid;

	struct xpc_partition *part;

	struct xpc_channel *ch;
@@ -793,10 +835,13 @@ xpc_disconnect_wait(int ch_number) 
		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);
@@ -806,62 +851,89 @@ xpc_disconnect_wait(int ch_number) 




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;



	/*

	 * 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);

	/* a 'rmmod XPC' and a 'reboot' cannot both end up here together */

	DBUG_ON(xpc_exiting == 1);



	/*

	 * 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) {

			if (xpc_partition_disengaged(part) &&

					part->act_state == XPC_P_INACTIVE) {

				continue;

			}



			active_part_count++;



				XPC_DEACTIVATE_PARTITION(part, xpcUnloading);

			XPC_DEACTIVATE_PARTITION(part, reason);

		}



		if (active_part_count == 0) {

			break;

		}



		if (active_part_count)

			msleep_interruptible(300);

	} while (active_part_count > 0);

		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);

		}



		/* 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();
@@ -876,6 +948,34 @@ xpc_do_exit(void) 
}





/*

 * 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)

{
@@ -920,6 +1020,12 @@ 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);
@@ -976,6 +1082,13 @@ xpc_init(void) 
	}





	/* 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
@@ -997,6 +1110,9 @@ xpc_init(void) 
		/* 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();
@@ -1018,9 +1134,9 @@ xpc_init(void) 
		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;

	}
@@ -1039,7 +1155,7 @@ module_init(xpc_init); 
void __exit

xpc_exit(void)

{

	xpc_do_exit();

	xpc_do_exit(xpcUnloading);

}

module_exit(xpc_exit);

arch/ia64/sn/kernel/xpc_partition.c

+254 −50

File changed.

Preview size limit exceeded, changes collapsed.

include/asm-ia64/sn/xp.h

+9 −1
Original line number Diff line number Diff line
@@ -217,7 +217,15 @@ enum xpc_retval { 
	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 */

};