2 * This file is subject to the terms and conditions of the GNU General Public
3 * License. See the file "COPYING" in the main directory of this archive
6 * Copyright (c) 2004-2008 Silicon Graphics, Inc. All Rights Reserved.
10 * Cross Partition Communication (XPC) structures and macros.
13 #ifndef _DRIVERS_MISC_SGIXP_XPC_H
14 #define _DRIVERS_MISC_SGIXP_XPC_H
16 #include <linux/interrupt.h>
17 #include <linux/sysctl.h>
18 #include <linux/device.h>
19 #include <linux/mutex.h>
20 #include <linux/completion.h>
21 #include <asm/pgtable.h>
22 #include <asm/processor.h>
23 #include <asm/sn/clksupport.h>
24 #include <asm/sn/addrs.h>
25 #include <asm/sn/mspec.h>
26 #include <asm/sn/shub_mmr.h>
30 * XPC Version numbers consist of a major and minor number. XPC can always
31 * talk to versions with same major #, and never talk to versions with a
34 #define _XPC_VERSION(_maj, _min) (((_maj) << 4) | ((_min) & 0xf))
35 #define XPC_VERSION_MAJOR(_v) ((_v) >> 4)
36 #define XPC_VERSION_MINOR(_v) ((_v) & 0xf)
39 * The next macros define word or bit representations for given
40 * C-brick nasid in either the SAL provided bit array representing
41 * nasids in the partition/machine or the AMO_t array used for
42 * inter-partition initiation communications.
44 * For SN2 machines, C-Bricks are alway even numbered NASIDs. As
45 * such, some space will be saved by insisting that nasid information
46 * passed from SAL always be packed for C-Bricks and the
47 * cross-partition interrupts use the same packing scheme.
49 #define XPC_NASID_W_INDEX(_n) (((_n) / 64) / 2)
50 #define XPC_NASID_B_INDEX(_n) (((_n) / 2) & (64 - 1))
51 #define XPC_NASID_IN_ARRAY(_n, _p) ((_p)[XPC_NASID_W_INDEX(_n)] & \
52 (1UL << XPC_NASID_B_INDEX(_n)))
53 #define XPC_NASID_FROM_W_B(_w, _b) (((_w) * 64 + (_b)) * 2)
55 #define XPC_HB_DEFAULT_INTERVAL 5 /* incr HB every x secs */
56 #define XPC_HB_CHECK_DEFAULT_INTERVAL 20 /* check HB every x secs */
58 /* define the process name of HB checker and the CPU it is pinned to */
59 #define XPC_HB_CHECK_THREAD_NAME "xpc_hb"
60 #define XPC_HB_CHECK_CPU 0
62 /* define the process name of the discovery thread */
63 #define XPC_DISCOVERY_THREAD_NAME "xpc_discovery"
68 * SAL reserves one page of memory per partition for XPC. Though a full page
69 * in length (16384 bytes), its starting address is not page aligned, but it
70 * is cacheline aligned. The reserved page consists of the following:
72 * reserved page header
74 * The first two 64-byte cachelines of the reserved page contain the
75 * header (struct xpc_rsvd_page). Before SAL initialization has completed,
76 * SAL has set up the following fields of the reserved page header:
77 * SAL_signature, SAL_version, SAL_partid, and SAL_nasids_size. The
78 * other fields are set up by XPC. (xpc_rsvd_page points to the local
79 * partition's reserved page.)
84 * SAL also sets up two bitmaps (or masks), one that reflects the actual
85 * nasids in this partition (part_nasids), and the other that reflects
86 * the actual nasids in the entire machine (mach_nasids). We're only
87 * interested in the even numbered nasids (which contain the processors
88 * and/or memory), so we only need half as many bits to represent the
89 * nasids. The part_nasids mask is located starting at the first cacheline
90 * following the reserved page header. The mach_nasids mask follows right
91 * after the part_nasids mask. The size in bytes of each mask is reflected
92 * by the reserved page header field 'SAL_nasids_size'. (Local partition's
93 * mask pointers are xpc_part_nasids and xpc_mach_nasids.)
95 * vars (ia64-sn2 only)
96 * vars part (ia64-sn2 only)
98 * Immediately following the mach_nasids mask are the XPC variables
99 * required by other partitions. First are those that are generic to all
100 * partitions (vars), followed on the next available cacheline by those
101 * which are partition specific (vars part). These are setup by XPC.
102 * (Local partition's vars pointers are xpc_vars and xpc_vars_part.)
104 * Note: Until 'stamp' is set non-zero, the partition XPC code has not been
107 struct xpc_rsvd_page
{
108 u64 SAL_signature
; /* SAL: unique signature */
109 u64 SAL_version
; /* SAL: version */
110 short SAL_partid
; /* SAL: partition ID */
111 short max_npartitions
; /* value of XPC_MAX_PARTITIONS */
113 u8 pad1
[3]; /* align to next u64 in 1st 64-byte cacheline */
115 u64 vars_pa
; /* physical address of struct xpc_vars */
116 u64 activate_mq_gpa
; /* global phys address of activate_mq */
118 struct timespec stamp
; /* time when reserved page was setup by XPC */
119 u64 pad2
[9]; /* align to last u64 in 2nd 64-byte cacheline */
120 u64 SAL_nasids_size
; /* SAL: size of each nasid mask in bytes */
123 #define XPC_RP_VERSION _XPC_VERSION(2, 0) /* version 2.0 of the reserved page */
125 #define XPC_SUPPORTS_RP_STAMP(_version) \
126 (_version >= _XPC_VERSION(1, 1))
128 #define ZERO_STAMP ((struct timespec){0, 0})
130 * compare stamps - the return value is:
132 * < 0, if stamp1 < stamp2
133 * = 0, if stamp1 == stamp2
134 * > 0, if stamp1 > stamp2
137 xpc_compare_stamps(struct timespec
*stamp1
, struct timespec
*stamp2
)
141 ret
= stamp1
->tv_sec
- stamp2
->tv_sec
;
143 ret
= stamp1
->tv_nsec
- stamp2
->tv_nsec
;
149 * Define the structures by which XPC variables can be exported to other
150 * partitions. (There are two: struct xpc_vars and struct xpc_vars_part)
154 * The following structure describes the partition generic variables
155 * needed by other partitions in order to properly initialize.
157 * struct xpc_vars version number also applies to struct xpc_vars_part.
158 * Changes to either structure and/or related functionality should be
159 * reflected by incrementing either the major or minor version numbers
160 * of struct xpc_vars.
165 u64 heartbeating_to_mask
;
166 u64 heartbeat_offline
; /* if 0, heartbeat should be changing */
170 u64 amos_page_pa
; /* paddr of page of AMOs from MSPEC driver */
171 AMO_t
*amos_page
; /* vaddr of page of AMOs from MSPEC driver */
174 #define XPC_V_VERSION _XPC_VERSION(3, 1) /* version 3.1 of the cross vars */
176 #define XPC_SUPPORTS_DISENGAGE_REQUEST(_version) \
177 (_version >= _XPC_VERSION(3, 1))
180 xpc_hb_allowed(short partid
, struct xpc_vars
*vars
)
182 return ((vars
->heartbeating_to_mask
& (1UL << partid
)) != 0);
186 xpc_allow_hb(short partid
, struct xpc_vars
*vars
)
188 u64 old_mask
, new_mask
;
191 old_mask
= vars
->heartbeating_to_mask
;
192 new_mask
= (old_mask
| (1UL << partid
));
193 } while (cmpxchg(&vars
->heartbeating_to_mask
, old_mask
, new_mask
) !=
198 xpc_disallow_hb(short partid
, struct xpc_vars
*vars
)
200 u64 old_mask
, new_mask
;
203 old_mask
= vars
->heartbeating_to_mask
;
204 new_mask
= (old_mask
& ~(1UL << partid
));
205 } while (cmpxchg(&vars
->heartbeating_to_mask
, old_mask
, new_mask
) !=
210 * The AMOs page consists of a number of AMO variables which are divided into
211 * four groups, The first two groups are used to identify an IRQ's sender.
212 * These two groups consist of 64 and 128 AMO variables respectively. The last
213 * two groups, consisting of just one AMO variable each, are used to identify
214 * the remote partitions that are currently engaged (from the viewpoint of
215 * the XPC running on the remote partition).
217 #define XPC_NOTIFY_IRQ_AMOS 0
218 #define XPC_ACTIVATE_IRQ_AMOS (XPC_NOTIFY_IRQ_AMOS + XP_MAX_NPARTITIONS_SN2)
219 #define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
220 #define XPC_DISENGAGE_REQUEST_AMO (XPC_ENGAGED_PARTITIONS_AMO + 1)
223 * The following structure describes the per partition specific variables.
225 * An array of these structures, one per partition, will be defined. As a
226 * partition becomes active XPC will copy the array entry corresponding to
227 * itself from that partition. It is desirable that the size of this structure
228 * evenly divides into a 128-byte cacheline, such that none of the entries in
229 * this array crosses a 128-byte cacheline boundary. As it is now, each entry
232 struct xpc_vars_part_sn2
{
235 u64 openclose_args_pa
; /* physical address of open and close args */
236 u64 GPs_pa
; /* physical address of Get/Put values */
238 u64 IPI_amo_pa
; /* physical address of IPI AMO_t structure */
239 int IPI_nasid
; /* nasid of where to send IPIs */
240 int IPI_phys_cpuid
; /* physical CPU ID of where to send IPIs */
242 u8 nchannels
; /* #of defined channels supported */
244 u8 reserved
[23]; /* pad to a full 64 bytes */
248 * The vars_part MAGIC numbers play a part in the first contact protocol.
250 * MAGIC1 indicates that the per partition specific variables for a remote
251 * partition have been initialized by this partition.
253 * MAGIC2 indicates that this partition has pulled the remote partititions
254 * per partition variables that pertain to this partition.
256 #define XPC_VP_MAGIC1 0x0053524156435058L /* 'XPCVARS\0'L (little endian) */
257 #define XPC_VP_MAGIC2 0x0073726176435058L /* 'XPCvars\0'L (little endian) */
259 /* the reserved page sizes and offsets */
261 #define XPC_RP_HEADER_SIZE L1_CACHE_ALIGN(sizeof(struct xpc_rsvd_page))
262 #define XPC_RP_VARS_SIZE L1_CACHE_ALIGN(sizeof(struct xpc_vars))
264 #define XPC_RP_PART_NASIDS(_rp) ((u64 *)((u8 *)(_rp) + XPC_RP_HEADER_SIZE))
265 #define XPC_RP_MACH_NASIDS(_rp) (XPC_RP_PART_NASIDS(_rp) + xp_nasid_mask_words)
266 #define XPC_RP_VARS(_rp) ((struct xpc_vars *)(XPC_RP_MACH_NASIDS(_rp) + \
267 xp_nasid_mask_words))
270 * Functions registered by add_timer() or called by kernel_thread() only
271 * allow for a single 64-bit argument. The following macros can be used to
272 * pack and unpack two (32-bit, 16-bit or 8-bit) arguments into or out from
273 * the passed argument.
275 #define XPC_PACK_ARGS(_arg1, _arg2) \
276 ((((u64) _arg1) & 0xffffffff) | \
277 ((((u64) _arg2) & 0xffffffff) << 32))
279 #define XPC_UNPACK_ARG1(_args) (((u64) _args) & 0xffffffff)
280 #define XPC_UNPACK_ARG2(_args) ((((u64) _args) >> 32) & 0xffffffff)
283 * Define a Get/Put value pair (pointers) used with a message queue.
286 s64 get
; /* Get value */
287 s64 put
; /* Put value */
290 #define XPC_GP_SIZE \
291 L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_MAX_NCHANNELS)
294 * Define a structure that contains arguments associated with opening and
297 struct xpc_openclose_args
{
298 u16 reason
; /* reason why channel is closing */
299 u16 msg_size
; /* sizeof each message entry */
300 u16 remote_nentries
; /* #of message entries in remote msg queue */
301 u16 local_nentries
; /* #of message entries in local msg queue */
302 u64 local_msgqueue_pa
; /* physical address of local message queue */
305 #define XPC_OPENCLOSE_ARGS_SIZE \
306 L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * \
309 /* struct xpc_msg flags */
311 #define XPC_M_DONE 0x01 /* msg has been received/consumed */
312 #define XPC_M_READY 0x02 /* msg is ready to be sent */
313 #define XPC_M_INTERRUPT 0x04 /* send interrupt when msg consumed */
315 #define XPC_MSG_ADDRESS(_payload) \
316 ((struct xpc_msg *)((u8 *)(_payload) - XPC_MSG_PAYLOAD_OFFSET))
319 * Defines notify entry.
321 * This is used to notify a message's sender that their message was received
322 * and consumed by the intended recipient.
325 u8 type
; /* type of notification */
327 /* the following two fields are only used if type == XPC_N_CALL */
328 xpc_notify_func func
; /* user's notify function */
329 void *key
; /* pointer to user's key */
332 /* struct xpc_notify type of notification */
334 #define XPC_N_CALL 0x01 /* notify function provided by user */
337 * Define the structure that manages all the stuff required by a channel. In
338 * particular, they are used to manage the messages sent across the channel.
340 * This structure is private to a partition, and is NOT shared across the
341 * partition boundary.
343 * There is an array of these structures for each remote partition. It is
344 * allocated at the time a partition becomes active. The array contains one
345 * of these structures for each potential channel connection to that partition.
347 * Each of these structures manages two message queues (circular buffers).
348 * They are allocated at the time a channel connection is made. One of
349 * these message queues (local_msgqueue) holds the locally created messages
350 * that are destined for the remote partition. The other of these message
351 * queues (remote_msgqueue) is a locally cached copy of the remote partition's
352 * own local_msgqueue.
354 * The following is a description of the Get/Put pointers used to manage these
355 * two message queues. Consider the local_msgqueue to be on one partition
356 * and the remote_msgqueue to be its cached copy on another partition. A
357 * description of what each of the lettered areas contains is included.
360 * local_msgqueue remote_msgqueue
362 * |/////////| |/////////|
363 * w_remote_GP.get --> +---------+ |/////////|
365 * remote_GP.get --> +---------+ +---------+ <-- local_GP->get
369 * | | +---------+ <-- w_local_GP.get
373 * | | +---------+ <-- w_remote_GP.put
375 * local_GP->put --> +---------+ +---------+ <-- remote_GP.put
379 * w_local_GP.put --> +---------+ |/////////|
380 * |/////////| |/////////|
383 * ( remote_GP.[get|put] are cached copies of the remote
384 * partition's local_GP->[get|put], and thus their values can
385 * lag behind their counterparts on the remote partition. )
388 * A - Messages that have been allocated, but have not yet been sent to the
391 * B - Messages that have been sent, but have not yet been acknowledged by the
392 * remote partition as having been received.
394 * C - Area that needs to be prepared for the copying of sent messages, by
395 * the clearing of the message flags of any previously received messages.
397 * D - Area into which sent messages are to be copied from the remote
398 * partition's local_msgqueue and then delivered to their intended
399 * recipients. [ To allow for a multi-message copy, another pointer
400 * (next_msg_to_pull) has been added to keep track of the next message
401 * number needing to be copied (pulled). It chases after w_remote_GP.put.
402 * Any messages lying between w_local_GP.get and next_msg_to_pull have
403 * been copied and are ready to be delivered. ]
405 * E - Messages that have been copied and delivered, but have not yet been
406 * acknowledged by the recipient as having been received.
408 * F - Messages that have been acknowledged, but XPC has not yet notified the
409 * sender that the message was received by its intended recipient.
410 * This is also an area that needs to be prepared for the allocating of
411 * new messages, by the clearing of the message flags of the acknowledged
415 short partid
; /* ID of remote partition connected */
416 spinlock_t lock
; /* lock for updating this structure */
417 u32 flags
; /* general flags */
419 enum xp_retval reason
; /* reason why channel is disconnect'g */
420 int reason_line
; /* line# disconnect initiated from */
422 u16 number
; /* channel # */
424 u16 msg_size
; /* sizeof each msg entry */
425 u16 local_nentries
; /* #of msg entries in local msg queue */
426 u16 remote_nentries
; /* #of msg entries in remote msg queue */
428 void *local_msgqueue_base
; /* base address of kmalloc'd space */
429 struct xpc_msg
*local_msgqueue
; /* local message queue */
430 void *remote_msgqueue_base
; /* base address of kmalloc'd space */
431 struct xpc_msg
*remote_msgqueue
; /* cached copy of remote partition's */
432 /* local message queue */
433 u64 remote_msgqueue_pa
; /* phys addr of remote partition's */
434 /* local message queue */
436 atomic_t references
; /* #of external references to queues */
438 atomic_t n_on_msg_allocate_wq
; /* #on msg allocation wait queue */
439 wait_queue_head_t msg_allocate_wq
; /* msg allocation wait queue */
441 u8 delayed_IPI_flags
; /* IPI flags received, but delayed */
442 /* action until channel disconnected */
444 /* queue of msg senders who want to be notified when msg received */
446 atomic_t n_to_notify
; /* #of msg senders to notify */
447 struct xpc_notify
*notify_queue
; /* notify queue for messages sent */
449 xpc_channel_func func
; /* user's channel function */
450 void *key
; /* pointer to user's key */
452 struct mutex msg_to_pull_mutex
; /* next msg to pull serialization */
453 struct completion wdisconnect_wait
; /* wait for channel disconnect */
455 struct xpc_openclose_args
*local_openclose_args
; /* args passed on */
456 /* opening or closing of channel */
458 /* various flavors of local and remote Get/Put values */
460 struct xpc_gp
*local_GP
; /* local Get/Put values */
461 struct xpc_gp remote_GP
; /* remote Get/Put values */
462 struct xpc_gp w_local_GP
; /* working local Get/Put values */
463 struct xpc_gp w_remote_GP
; /* working remote Get/Put values */
464 s64 next_msg_to_pull
; /* Put value of next msg to pull */
466 /* kthread management related fields */
468 atomic_t kthreads_assigned
; /* #of kthreads assigned to channel */
469 u32 kthreads_assigned_limit
; /* limit on #of kthreads assigned */
470 atomic_t kthreads_idle
; /* #of kthreads idle waiting for work */
471 u32 kthreads_idle_limit
; /* limit on #of kthreads idle */
472 atomic_t kthreads_active
; /* #of kthreads actively working */
474 wait_queue_head_t idle_wq
; /* idle kthread wait queue */
476 } ____cacheline_aligned
;
478 /* struct xpc_channel flags */
480 #define XPC_C_WASCONNECTED 0x00000001 /* channel was connected */
482 #define XPC_C_ROPENREPLY 0x00000002 /* remote open channel reply */
483 #define XPC_C_OPENREPLY 0x00000004 /* local open channel reply */
484 #define XPC_C_ROPENREQUEST 0x00000008 /* remote open channel request */
485 #define XPC_C_OPENREQUEST 0x00000010 /* local open channel request */
487 #define XPC_C_SETUP 0x00000020 /* channel's msgqueues are alloc'd */
488 #define XPC_C_CONNECTEDCALLOUT 0x00000040 /* connected callout initiated */
489 #define XPC_C_CONNECTEDCALLOUT_MADE \
490 0x00000080 /* connected callout completed */
491 #define XPC_C_CONNECTED 0x00000100 /* local channel is connected */
492 #define XPC_C_CONNECTING 0x00000200 /* channel is being connected */
494 #define XPC_C_RCLOSEREPLY 0x00000400 /* remote close channel reply */
495 #define XPC_C_CLOSEREPLY 0x00000800 /* local close channel reply */
496 #define XPC_C_RCLOSEREQUEST 0x00001000 /* remote close channel request */
497 #define XPC_C_CLOSEREQUEST 0x00002000 /* local close channel request */
499 #define XPC_C_DISCONNECTED 0x00004000 /* channel is disconnected */
500 #define XPC_C_DISCONNECTING 0x00008000 /* channel is being disconnected */
501 #define XPC_C_DISCONNECTINGCALLOUT \
502 0x00010000 /* disconnecting callout initiated */
503 #define XPC_C_DISCONNECTINGCALLOUT_MADE \
504 0x00020000 /* disconnecting callout completed */
505 #define XPC_C_WDISCONNECT 0x00040000 /* waiting for channel disconnect */
508 * Manages channels on a partition basis. There is one of these structures
509 * for each partition (a partition will never utilize the structure that
510 * represents itself).
512 struct xpc_partition
{
514 /* XPC HB infrastructure */
516 u8 remote_rp_version
; /* version# of partition's rsvd pg */
517 struct timespec remote_rp_stamp
; /* time when rsvd pg was initialized */
518 u64 remote_rp_pa
; /* phys addr of partition's rsvd pg */
519 u64 remote_vars_pa
; /* phys addr of partition's vars */
520 u64 remote_vars_part_pa
; /* phys addr of partition's vars part */
521 u64 last_heartbeat
; /* HB at last read */
522 u64 remote_amos_page_pa
; /* phys addr of partition's amos page */
523 int remote_act_nasid
; /* active part's act/deact nasid */
524 int remote_act_phys_cpuid
; /* active part's act/deact phys cpuid */
525 u32 act_IRQ_rcvd
; /* IRQs since activation */
526 spinlock_t act_lock
; /* protect updating of act_state */
527 u8 act_state
; /* from XPC HB viewpoint */
528 u8 remote_vars_version
; /* version# of partition's vars */
529 enum xp_retval reason
; /* reason partition is deactivating */
530 int reason_line
; /* line# deactivation initiated from */
531 int reactivate_nasid
; /* nasid in partition to reactivate */
533 unsigned long disengage_request_timeout
; /* timeout in jiffies */
534 struct timer_list disengage_request_timer
;
536 /* XPC infrastructure referencing and teardown control */
538 u8 setup_state
; /* infrastructure setup state */
539 wait_queue_head_t teardown_wq
; /* kthread waiting to teardown infra */
540 atomic_t references
; /* #of references to infrastructure */
542 u8 nchannels
; /* #of defined channels supported */
543 atomic_t nchannels_active
; /* #of channels that are not DISCONNECTED */
544 atomic_t nchannels_engaged
; /* #of channels engaged with remote part */
545 struct xpc_channel
*channels
; /* array of channel structures */
547 void *local_GPs_base
; /* base address of kmalloc'd space */
548 struct xpc_gp
*local_GPs
; /* local Get/Put values */
549 void *remote_GPs_base
; /* base address of kmalloc'd space */
550 struct xpc_gp
*remote_GPs
; /* copy of remote partition's local */
552 u64 remote_GPs_pa
; /* phys address of remote partition's local */
555 /* fields used to pass args when opening or closing a channel */
557 void *local_openclose_args_base
; /* base address of kmalloc'd space */
558 struct xpc_openclose_args
*local_openclose_args
; /* local's args */
559 void *remote_openclose_args_base
; /* base address of kmalloc'd space */
560 struct xpc_openclose_args
*remote_openclose_args
; /* copy of remote's */
562 u64 remote_openclose_args_pa
; /* phys addr of remote's args */
564 /* IPI sending, receiving and handling related fields */
566 int remote_IPI_nasid
; /* nasid of where to send IPIs */
567 int remote_IPI_phys_cpuid
; /* phys CPU ID of where to send IPIs */
568 AMO_t
*remote_IPI_amo_va
; /* address of remote IPI AMO_t structure */
570 AMO_t
*local_IPI_amo_va
; /* address of IPI AMO_t structure */
571 u64 local_IPI_amo
; /* IPI amo flags yet to be handled */
572 char IPI_owner
[8]; /* IPI owner's name */
573 struct timer_list dropped_IPI_timer
; /* dropped IPI timer */
575 spinlock_t IPI_lock
; /* IPI handler lock */
577 /* channel manager related fields */
579 atomic_t channel_mgr_requests
; /* #of requests to activate chan mgr */
580 wait_queue_head_t channel_mgr_wq
; /* channel mgr's wait queue */
582 } ____cacheline_aligned
;
584 /* struct xpc_partition act_state values (for XPC HB) */
586 #define XPC_P_INACTIVE 0x00 /* partition is not active */
587 #define XPC_P_ACTIVATION_REQ 0x01 /* created thread to activate */
588 #define XPC_P_ACTIVATING 0x02 /* activation thread started */
589 #define XPC_P_ACTIVE 0x03 /* xpc_partition_up() was called */
590 #define XPC_P_DEACTIVATING 0x04 /* partition deactivation initiated */
592 #define XPC_DEACTIVATE_PARTITION(_p, _reason) \
593 xpc_deactivate_partition(__LINE__, (_p), (_reason))
595 /* struct xpc_partition setup_state values */
597 #define XPC_P_UNSET 0x00 /* infrastructure was never setup */
598 #define XPC_P_SETUP 0x01 /* infrastructure is setup */
599 #define XPC_P_WTEARDOWN 0x02 /* waiting to teardown infrastructure */
600 #define XPC_P_TORNDOWN 0x03 /* infrastructure is torndown */
603 * struct xpc_partition IPI_timer #of seconds to wait before checking for
604 * dropped IPIs. These occur whenever an IPI amo write doesn't complete until
605 * after the IPI was received.
607 #define XPC_P_DROPPED_IPI_WAIT_INTERVAL (0.25 * HZ)
609 /* number of seconds to wait for other partitions to disengage */
610 #define XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT 90
612 /* interval in seconds to print 'waiting disengagement' messages */
613 #define XPC_DISENGAGE_PRINTMSG_INTERVAL 10
615 #define XPC_PARTID(_p) ((short)((_p) - &xpc_partitions[0]))
617 /* found in xp_main.c */
618 extern struct xpc_registration xpc_registrations
[];
620 /* found in xpc_main.c */
621 extern struct device
*xpc_part
;
622 extern struct device
*xpc_chan
;
623 extern int xpc_disengage_request_timelimit
;
624 extern int xpc_disengage_request_timedout
;
625 extern irqreturn_t
xpc_notify_IRQ_handler(int, void *);
626 extern void xpc_dropped_IPI_check(struct xpc_partition
*);
627 extern void xpc_activate_partition(struct xpc_partition
*);
628 extern void xpc_activate_kthreads(struct xpc_channel
*, int);
629 extern void xpc_create_kthreads(struct xpc_channel
*, int, int);
630 extern void xpc_disconnect_wait(int);
631 extern enum xp_retval (*xpc_rsvd_page_init
) (struct xpc_rsvd_page
*);
632 extern enum xp_retval (*xpc_make_first_contact
) (struct xpc_partition
*);
633 extern u64 (*xpc_get_IPI_flags
) (struct xpc_partition
*);
634 extern struct xpc_msg
*(*xpc_get_deliverable_msg
) (struct xpc_channel
*);
635 extern enum xp_retval (*xpc_setup_infrastructure
) (struct xpc_partition
*);
636 extern void (*xpc_teardown_infrastructure
) (struct xpc_partition
*);
638 /* found in xpc_sn2.c */
639 extern void xpc_init_sn2(void);
640 extern struct xpc_vars
*xpc_vars
; /*>>> eliminate from here */
642 /* found in xpc_uv.c */
643 extern void xpc_init_uv(void);
645 /* found in xpc_partition.c */
646 extern int xpc_exiting
;
647 extern int xp_nasid_mask_words
;
648 extern struct xpc_rsvd_page
*xpc_rsvd_page
;
649 extern struct xpc_partition
*xpc_partitions
;
650 extern char *xpc_remote_copy_buffer
;
651 extern void *xpc_remote_copy_buffer_base
;
652 extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t
, void **);
653 extern struct xpc_rsvd_page
*xpc_setup_rsvd_page(void);
654 extern void xpc_allow_IPI_ops(void);
655 extern void xpc_restrict_IPI_ops(void);
656 extern int xpc_identify_act_IRQ_sender(void);
657 extern int xpc_partition_disengaged(struct xpc_partition
*);
658 extern enum xp_retval
xpc_mark_partition_active(struct xpc_partition
*);
659 extern void xpc_mark_partition_inactive(struct xpc_partition
*);
660 extern void xpc_discovery(void);
661 extern void xpc_check_remote_hb(void);
662 extern void xpc_deactivate_partition(const int, struct xpc_partition
*,
664 extern enum xp_retval
xpc_initiate_partid_to_nasids(short, void *);
666 /* found in xpc_channel.c */
667 extern void *xpc_kzalloc_cacheline_aligned(size_t, gfp_t
, void **);
668 extern void xpc_initiate_connect(int);
669 extern void xpc_initiate_disconnect(int);
670 extern enum xp_retval
xpc_initiate_allocate(short, int, u32
, void **);
671 extern enum xp_retval
xpc_initiate_send(short, int, void *);
672 extern enum xp_retval
xpc_initiate_send_notify(short, int, void *,
673 xpc_notify_func
, void *);
674 extern void xpc_initiate_received(short, int, void *);
675 extern void xpc_process_channel_activity(struct xpc_partition
*);
676 extern void xpc_connected_callout(struct xpc_channel
*);
677 extern void xpc_deliver_msg(struct xpc_channel
*);
678 extern void xpc_disconnect_channel(const int, struct xpc_channel
*,
679 enum xp_retval
, unsigned long *);
680 extern void xpc_disconnect_callout(struct xpc_channel
*, enum xp_retval
);
681 extern void xpc_partition_going_down(struct xpc_partition
*, enum xp_retval
);
684 xpc_wakeup_channel_mgr(struct xpc_partition
*part
)
686 if (atomic_inc_return(&part
->channel_mgr_requests
) == 1)
687 wake_up(&part
->channel_mgr_wq
);
691 * These next two inlines are used to keep us from tearing down a channel's
692 * msg queues while a thread may be referencing them.
695 xpc_msgqueue_ref(struct xpc_channel
*ch
)
697 atomic_inc(&ch
->references
);
701 xpc_msgqueue_deref(struct xpc_channel
*ch
)
703 s32 refs
= atomic_dec_return(&ch
->references
);
707 xpc_wakeup_channel_mgr(&xpc_partitions
[ch
->partid
]);
710 #define XPC_DISCONNECT_CHANNEL(_ch, _reason, _irqflgs) \
711 xpc_disconnect_channel(__LINE__, _ch, _reason, _irqflgs)
714 * These two inlines are used to keep us from tearing down a partition's
715 * setup infrastructure while a thread may be referencing it.
718 xpc_part_deref(struct xpc_partition
*part
)
720 s32 refs
= atomic_dec_return(&part
->references
);
723 if (refs
== 0 && part
->setup_state
== XPC_P_WTEARDOWN
)
724 wake_up(&part
->teardown_wq
);
728 xpc_part_ref(struct xpc_partition
*part
)
732 atomic_inc(&part
->references
);
733 setup
= (part
->setup_state
== XPC_P_SETUP
);
735 xpc_part_deref(part
);
741 * The following macro is to be used for the setting of the reason and
742 * reason_line fields in both the struct xpc_channel and struct xpc_partition
745 #define XPC_SET_REASON(_p, _reason, _line) \
747 (_p)->reason = _reason; \
748 (_p)->reason_line = _line; \
752 * This next set of inlines are used to keep track of when a partition is
753 * potentially engaged in accessing memory belonging to another partition.
757 xpc_mark_partition_engaged(struct xpc_partition
*part
)
759 unsigned long irq_flags
;
760 AMO_t
*amo
= (AMO_t
*)__va(part
->remote_amos_page_pa
+
761 (XPC_ENGAGED_PARTITIONS_AMO
*
764 local_irq_save(irq_flags
);
766 /* set bit corresponding to our partid in remote partition's AMO */
767 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_OR
,
768 (1UL << sn_partition_id
));
770 * We must always use the nofault function regardless of whether we
771 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
772 * didn't, we'd never know that the other partition is down and would
773 * keep sending IPIs and AMOs to it until the heartbeat times out.
775 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
777 xp_nofault_PIOR_target
));
779 local_irq_restore(irq_flags
);
783 xpc_mark_partition_disengaged(struct xpc_partition
*part
)
785 unsigned long irq_flags
;
786 AMO_t
*amo
= (AMO_t
*)__va(part
->remote_amos_page_pa
+
787 (XPC_ENGAGED_PARTITIONS_AMO
*
790 local_irq_save(irq_flags
);
792 /* clear bit corresponding to our partid in remote partition's AMO */
793 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
794 ~(1UL << sn_partition_id
));
796 * We must always use the nofault function regardless of whether we
797 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
798 * didn't, we'd never know that the other partition is down and would
799 * keep sending IPIs and AMOs to it until the heartbeat times out.
801 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
803 xp_nofault_PIOR_target
));
805 local_irq_restore(irq_flags
);
809 xpc_request_partition_disengage(struct xpc_partition
*part
)
811 unsigned long irq_flags
;
812 AMO_t
*amo
= (AMO_t
*)__va(part
->remote_amos_page_pa
+
813 (XPC_DISENGAGE_REQUEST_AMO
* sizeof(AMO_t
)));
815 local_irq_save(irq_flags
);
817 /* set bit corresponding to our partid in remote partition's AMO */
818 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_OR
,
819 (1UL << sn_partition_id
));
821 * We must always use the nofault function regardless of whether we
822 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
823 * didn't, we'd never know that the other partition is down and would
824 * keep sending IPIs and AMOs to it until the heartbeat times out.
826 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
828 xp_nofault_PIOR_target
));
830 local_irq_restore(irq_flags
);
834 xpc_cancel_partition_disengage_request(struct xpc_partition
*part
)
836 unsigned long irq_flags
;
837 AMO_t
*amo
= (AMO_t
*)__va(part
->remote_amos_page_pa
+
838 (XPC_DISENGAGE_REQUEST_AMO
* sizeof(AMO_t
)));
840 local_irq_save(irq_flags
);
842 /* clear bit corresponding to our partid in remote partition's AMO */
843 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
844 ~(1UL << sn_partition_id
));
846 * We must always use the nofault function regardless of whether we
847 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
848 * didn't, we'd never know that the other partition is down and would
849 * keep sending IPIs and AMOs to it until the heartbeat times out.
851 (void)xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->
853 xp_nofault_PIOR_target
));
855 local_irq_restore(irq_flags
);
859 xpc_partition_engaged(u64 partid_mask
)
861 AMO_t
*amo
= xpc_vars
->amos_page
+ XPC_ENGAGED_PARTITIONS_AMO
;
863 /* return our partition's AMO variable ANDed with partid_mask */
864 return (FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_LOAD
) &
869 xpc_partition_disengage_requested(u64 partid_mask
)
871 AMO_t
*amo
= xpc_vars
->amos_page
+ XPC_DISENGAGE_REQUEST_AMO
;
873 /* return our partition's AMO variable ANDed with partid_mask */
874 return (FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_LOAD
) &
879 xpc_clear_partition_engaged(u64 partid_mask
)
881 AMO_t
*amo
= xpc_vars
->amos_page
+ XPC_ENGAGED_PARTITIONS_AMO
;
883 /* clear bit(s) based on partid_mask in our partition's AMO */
884 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
889 xpc_clear_partition_disengage_request(u64 partid_mask
)
891 AMO_t
*amo
= xpc_vars
->amos_page
+ XPC_DISENGAGE_REQUEST_AMO
;
893 /* clear bit(s) based on partid_mask in our partition's AMO */
894 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_AND
,
899 * The following set of macros and inlines are used for the sending and
900 * receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
901 * one that is associated with partition activity (SGI_XPC_ACTIVATE) and
902 * the other that is associated with channel activity (SGI_XPC_NOTIFY).
906 xpc_IPI_receive(AMO_t
*amo
)
908 return FETCHOP_LOAD_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_CLEAR
);
911 static inline enum xp_retval
912 xpc_IPI_send(AMO_t
*amo
, u64 flag
, int nasid
, int phys_cpuid
, int vector
)
915 unsigned long irq_flags
;
917 local_irq_save(irq_flags
);
919 FETCHOP_STORE_OP(TO_AMO((u64
)&amo
->variable
), FETCHOP_OR
, flag
);
920 sn_send_IPI_phys(nasid
, phys_cpuid
, vector
, 0);
923 * We must always use the nofault function regardless of whether we
924 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
925 * didn't, we'd never know that the other partition is down and would
926 * keep sending IPIs and AMOs to it until the heartbeat times out.
928 ret
= xp_nofault_PIOR((u64
*)GLOBAL_MMR_ADDR(NASID_GET(&amo
->variable
),
929 xp_nofault_PIOR_target
));
931 local_irq_restore(irq_flags
);
933 return ((ret
== 0) ? xpSuccess
: xpPioReadError
);
937 * IPIs associated with SGI_XPC_ACTIVATE IRQ.
941 * Flag the appropriate AMO variable and send an IPI to the specified node.
944 xpc_activate_IRQ_send(u64 amos_page_pa
, int from_nasid
, int to_nasid
,
947 int w_index
= XPC_NASID_W_INDEX(from_nasid
);
948 int b_index
= XPC_NASID_B_INDEX(from_nasid
);
949 AMO_t
*amos
= (AMO_t
*)__va(amos_page_pa
+
950 (XPC_ACTIVATE_IRQ_AMOS
* sizeof(AMO_t
)));
952 (void)xpc_IPI_send(&amos
[w_index
], (1UL << b_index
), to_nasid
,
953 to_phys_cpuid
, SGI_XPC_ACTIVATE
);
957 xpc_IPI_send_activate(struct xpc_vars
*vars
)
959 xpc_activate_IRQ_send(vars
->amos_page_pa
, cnodeid_to_nasid(0),
960 vars
->act_nasid
, vars
->act_phys_cpuid
);
964 xpc_IPI_send_activated(struct xpc_partition
*part
)
966 xpc_activate_IRQ_send(part
->remote_amos_page_pa
, cnodeid_to_nasid(0),
967 part
->remote_act_nasid
,
968 part
->remote_act_phys_cpuid
);
972 xpc_IPI_send_reactivate(struct xpc_partition
*part
)
974 xpc_activate_IRQ_send(xpc_vars
->amos_page_pa
, part
->reactivate_nasid
,
975 xpc_vars
->act_nasid
, xpc_vars
->act_phys_cpuid
);
979 xpc_IPI_send_disengage(struct xpc_partition
*part
)
981 xpc_activate_IRQ_send(part
->remote_amos_page_pa
, cnodeid_to_nasid(0),
982 part
->remote_act_nasid
,
983 part
->remote_act_phys_cpuid
);
987 * IPIs associated with SGI_XPC_NOTIFY IRQ.
991 * Send an IPI to the remote partition that is associated with the
994 #define XPC_NOTIFY_IRQ_SEND(_ch, _ipi_f, _irq_f) \
995 xpc_notify_IRQ_send(_ch, _ipi_f, #_ipi_f, _irq_f)
998 xpc_notify_IRQ_send(struct xpc_channel
*ch
, u8 ipi_flag
, char *ipi_flag_string
,
999 unsigned long *irq_flags
)
1001 struct xpc_partition
*part
= &xpc_partitions
[ch
->partid
];
1004 if (likely(part
->act_state
!= XPC_P_DEACTIVATING
)) {
1005 ret
= xpc_IPI_send(part
->remote_IPI_amo_va
,
1006 (u64
)ipi_flag
<< (ch
->number
* 8),
1007 part
->remote_IPI_nasid
,
1008 part
->remote_IPI_phys_cpuid
, SGI_XPC_NOTIFY
);
1009 dev_dbg(xpc_chan
, "%s sent to partid=%d, channel=%d, ret=%d\n",
1010 ipi_flag_string
, ch
->partid
, ch
->number
, ret
);
1011 if (unlikely(ret
!= xpSuccess
)) {
1012 if (irq_flags
!= NULL
)
1013 spin_unlock_irqrestore(&ch
->lock
, *irq_flags
);
1014 XPC_DEACTIVATE_PARTITION(part
, ret
);
1015 if (irq_flags
!= NULL
)
1016 spin_lock_irqsave(&ch
->lock
, *irq_flags
);
1022 * Make it look like the remote partition, which is associated with the
1023 * specified channel, sent us an IPI. This faked IPI will be handled
1024 * by xpc_dropped_IPI_check().
1026 #define XPC_NOTIFY_IRQ_SEND_LOCAL(_ch, _ipi_f) \
1027 xpc_notify_IRQ_send_local(_ch, _ipi_f, #_ipi_f)
1030 xpc_notify_IRQ_send_local(struct xpc_channel
*ch
, u8 ipi_flag
,
1031 char *ipi_flag_string
)
1033 struct xpc_partition
*part
= &xpc_partitions
[ch
->partid
];
1035 FETCHOP_STORE_OP(TO_AMO((u64
)&part
->local_IPI_amo_va
->variable
),
1036 FETCHOP_OR
, ((u64
)ipi_flag
<< (ch
->number
* 8)));
1037 dev_dbg(xpc_chan
, "%s sent local from partid=%d, channel=%d\n",
1038 ipi_flag_string
, ch
->partid
, ch
->number
);
1042 * The sending and receiving of IPIs includes the setting of an AMO variable
1043 * to indicate the reason the IPI was sent. The 64-bit variable is divided
1044 * up into eight bytes, ordered from right to left. Byte zero pertains to
1045 * channel 0, byte one to channel 1, and so on. Each byte is described by
1046 * the following IPI flags.
1049 #define XPC_IPI_CLOSEREQUEST 0x01
1050 #define XPC_IPI_CLOSEREPLY 0x02
1051 #define XPC_IPI_OPENREQUEST 0x04
1052 #define XPC_IPI_OPENREPLY 0x08
1053 #define XPC_IPI_MSGREQUEST 0x10
1055 /* given an AMO variable and a channel#, get its associated IPI flags */
1056 #define XPC_GET_IPI_FLAGS(_amo, _c) ((u8) (((_amo) >> ((_c) * 8)) & 0xff))
1057 #define XPC_SET_IPI_FLAGS(_amo, _c, _f) (_amo) |= ((u64) (_f) << ((_c) * 8))
1059 #define XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(_amo) ((_amo) & 0x0f0f0f0f0f0f0f0fUL)
1060 #define XPC_ANY_MSG_IPI_FLAGS_SET(_amo) ((_amo) & 0x1010101010101010UL)
1063 xpc_IPI_send_closerequest(struct xpc_channel
*ch
, unsigned long *irq_flags
)
1065 struct xpc_openclose_args
*args
= ch
->local_openclose_args
;
1067 args
->reason
= ch
->reason
;
1069 XPC_NOTIFY_IRQ_SEND(ch
, XPC_IPI_CLOSEREQUEST
, irq_flags
);
1073 xpc_IPI_send_closereply(struct xpc_channel
*ch
, unsigned long *irq_flags
)
1075 XPC_NOTIFY_IRQ_SEND(ch
, XPC_IPI_CLOSEREPLY
, irq_flags
);
1079 xpc_IPI_send_openrequest(struct xpc_channel
*ch
, unsigned long *irq_flags
)
1081 struct xpc_openclose_args
*args
= ch
->local_openclose_args
;
1083 args
->msg_size
= ch
->msg_size
;
1084 args
->local_nentries
= ch
->local_nentries
;
1086 XPC_NOTIFY_IRQ_SEND(ch
, XPC_IPI_OPENREQUEST
, irq_flags
);
1090 xpc_IPI_send_openreply(struct xpc_channel
*ch
, unsigned long *irq_flags
)
1092 struct xpc_openclose_args
*args
= ch
->local_openclose_args
;
1094 args
->remote_nentries
= ch
->remote_nentries
;
1095 args
->local_nentries
= ch
->local_nentries
;
1096 args
->local_msgqueue_pa
= __pa(ch
->local_msgqueue
);
1098 XPC_NOTIFY_IRQ_SEND(ch
, XPC_IPI_OPENREPLY
, irq_flags
);
1102 xpc_IPI_send_msgrequest(struct xpc_channel
*ch
)
1104 XPC_NOTIFY_IRQ_SEND(ch
, XPC_IPI_MSGREQUEST
, NULL
);
1108 xpc_IPI_send_local_msgrequest(struct xpc_channel
*ch
)
1110 XPC_NOTIFY_IRQ_SEND_LOCAL(ch
, XPC_IPI_MSGREQUEST
);
1114 >>> this block comment needs to be moved and re-written.
1115 * Memory for XPC's AMO variables is allocated by the MSPEC driver. These
1116 * pages are located in the lowest granule. The lowest granule uses 4k pages
1117 * for cached references and an alternate TLB handler to never provide a
1118 * cacheable mapping for the entire region. This will prevent speculative
1119 * reading of cached copies of our lines from being issued which will cause
1120 * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
1121 * AMO variables (based on xp_max_npartitions) for message notification and an
1122 * additional 128 AMO variables (based on XP_NASID_MASK_WORDS) for partition
1123 * activation and 2 AMO variables for partition deactivation.
1125 static inline AMO_t
*
1126 xpc_IPI_init(int index
)
1128 AMO_t
*amo
= xpc_vars
->amos_page
+ index
;
1130 (void)xpc_IPI_receive(amo
); /* clear AMO variable */
1135 * Check to see if there is any channel activity to/from the specified
1139 xpc_check_for_channel_activity(struct xpc_partition
*part
)
1142 unsigned long irq_flags
;
1144 IPI_amo
= xpc_IPI_receive(part
->local_IPI_amo_va
);
1148 spin_lock_irqsave(&part
->IPI_lock
, irq_flags
);
1149 part
->local_IPI_amo
|= IPI_amo
;
1150 spin_unlock_irqrestore(&part
->IPI_lock
, irq_flags
);
1152 dev_dbg(xpc_chan
, "received IPI from partid=%d, IPI_amo=0x%lx\n",
1153 XPC_PARTID(part
), IPI_amo
);
1155 xpc_wakeup_channel_mgr(part
);
1158 #endif /* _DRIVERS_MISC_SGIXP_XPC_H */