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/bte.h>
24 #include <asm/sn/clksupport.h>
25 #include <asm/sn/addrs.h>
26 #include <asm/sn/mspec.h>
27 #include <asm/sn/shub_mmr.h>
31 * XPC Version numbers consist of a major and minor number. XPC can always
32 * talk to versions with same major #, and never talk to versions with a
35 #define _XPC_VERSION(_maj, _min) (((_maj) << 4) | ((_min) & 0xf))
36 #define XPC_VERSION_MAJOR(_v) ((_v) >> 4)
37 #define XPC_VERSION_MINOR(_v) ((_v) & 0xf)
40 * The next macros define word or bit representations for given
41 * C-brick nasid in either the SAL provided bit array representing
42 * nasids in the partition/machine or the AMO_t array used for
43 * inter-partition initiation communications.
45 * For SN2 machines, C-Bricks are alway even numbered NASIDs. As
46 * such, some space will be saved by insisting that nasid information
47 * passed from SAL always be packed for C-Bricks and the
48 * cross-partition interrupts use the same packing scheme.
50 #define XPC_NASID_W_INDEX(_n) (((_n) / 64) / 2)
51 #define XPC_NASID_B_INDEX(_n) (((_n) / 2) & (64 - 1))
52 #define XPC_NASID_IN_ARRAY(_n, _p) ((_p)[XPC_NASID_W_INDEX(_n)] & \
53 (1UL << XPC_NASID_B_INDEX(_n)))
54 #define XPC_NASID_FROM_W_B(_w, _b) (((_w) * 64 + (_b)) * 2)
56 #define XPC_HB_DEFAULT_INTERVAL 5 /* incr HB every x secs */
57 #define XPC_HB_CHECK_DEFAULT_INTERVAL 20 /* check HB every x secs */
59 /* define the process name of HB checker and the CPU it is pinned to */
60 #define XPC_HB_CHECK_THREAD_NAME "xpc_hb"
61 #define XPC_HB_CHECK_CPU 0
63 /* define the process name of the discovery thread */
64 #define XPC_DISCOVERY_THREAD_NAME "xpc_discovery"
69 * SAL reserves one page of memory per partition for XPC. Though a full page
70 * in length (16384 bytes), its starting address is not page aligned, but it
71 * is cacheline aligned. The reserved page consists of the following:
73 * reserved page header
75 * The first cacheline of the reserved page contains the header
76 * (struct xpc_rsvd_page). Before SAL initialization has completed,
77 * SAL has set up the following fields of the reserved page header:
78 * SAL_signature, SAL_version, partid, and nasids_size. The other
79 * fields are set up by XPC. (xpc_rsvd_page points to the local
80 * partition's reserved page.)
85 * SAL also sets up two bitmaps (or masks), one that reflects the actual
86 * nasids in this partition (part_nasids), and the other that reflects
87 * the actual nasids in the entire machine (mach_nasids). We're only
88 * interested in the even numbered nasids (which contain the processors
89 * and/or memory), so we only need half as many bits to represent the
90 * nasids. The part_nasids mask is located starting at the first cacheline
91 * following the reserved page header. The mach_nasids mask follows right
92 * after the part_nasids mask. The size in bytes of each mask is reflected
93 * by the reserved page header field 'nasids_size'. (Local partition's
94 * mask pointers are xpc_part_nasids and xpc_mach_nasids.)
99 * Immediately following the mach_nasids mask are the XPC variables
100 * required by other partitions. First are those that are generic to all
101 * partitions (vars), followed on the next available cacheline by those
102 * which are partition specific (vars part). These are setup by XPC.
103 * (Local partition's vars pointers are xpc_vars and xpc_vars_part.)
105 * Note: Until vars_pa is set, the partition XPC code has not been initialized.
107 struct xpc_rsvd_page {
108 u64 SAL_signature; /* SAL: unique signature */
109 u64 SAL_version; /* SAL: version */
110 u8 partid; /* SAL: partition ID */
112 u8 pad1[6]; /* align to next u64 in cacheline */
113 u64 vars_pa; /* physical address of struct xpc_vars */
114 struct timespec stamp; /* time when reserved page was setup by XPC */
115 u64 pad2[9]; /* align to last u64 in cacheline */
116 u64 nasids_size; /* SAL: size of each nasid mask in bytes */
119 #define XPC_RP_VERSION _XPC_VERSION(1, 1) /* version 1.1 of the reserved page */
121 #define XPC_SUPPORTS_RP_STAMP(_version) \
122 (_version >= _XPC_VERSION(1, 1))
125 * compare stamps - the return value is:
127 * < 0, if stamp1 < stamp2
128 * = 0, if stamp1 == stamp2
129 * > 0, if stamp1 > stamp2
132 xpc_compare_stamps(struct timespec *stamp1, struct timespec *stamp2)
136 ret = stamp1->tv_sec - stamp2->tv_sec;
138 ret = stamp1->tv_nsec - stamp2->tv_nsec;
144 * Define the structures by which XPC variables can be exported to other
145 * partitions. (There are two: struct xpc_vars and struct xpc_vars_part)
149 * The following structure describes the partition generic variables
150 * needed by other partitions in order to properly initialize.
152 * struct xpc_vars version number also applies to struct xpc_vars_part.
153 * Changes to either structure and/or related functionality should be
154 * reflected by incrementing either the major or minor version numbers
155 * of struct xpc_vars.
160 u64 heartbeating_to_mask;
161 u64 heartbeat_offline; /* if 0, heartbeat should be changing */
165 u64 amos_page_pa; /* paddr of page of AMOs from MSPEC driver */
166 AMO_t *amos_page; /* vaddr of page of AMOs from MSPEC driver */
169 #define XPC_V_VERSION _XPC_VERSION(3, 1) /* version 3.1 of the cross vars */
171 #define XPC_SUPPORTS_DISENGAGE_REQUEST(_version) \
172 (_version >= _XPC_VERSION(3, 1))
175 xpc_hb_allowed(short partid, struct xpc_vars *vars)
177 return ((vars->heartbeating_to_mask & (1UL << partid)) != 0);
181 xpc_allow_hb(short partid, struct xpc_vars *vars)
183 u64 old_mask, new_mask;
186 old_mask = vars->heartbeating_to_mask;
187 new_mask = (old_mask | (1UL << partid));
188 } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
193 xpc_disallow_hb(short partid, struct xpc_vars *vars)
195 u64 old_mask, new_mask;
198 old_mask = vars->heartbeating_to_mask;
199 new_mask = (old_mask & ~(1UL << partid));
200 } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
205 * The AMOs page consists of a number of AMO variables which are divided into
206 * four groups, The first two groups are used to identify an IRQ's sender.
207 * These two groups consist of 64 and 128 AMO variables respectively. The last
208 * two groups, consisting of just one AMO variable each, are used to identify
209 * the remote partitions that are currently engaged (from the viewpoint of
210 * the XPC running on the remote partition).
212 #define XPC_NOTIFY_IRQ_AMOS 0
213 #define XPC_ACTIVATE_IRQ_AMOS (XPC_NOTIFY_IRQ_AMOS + XP_MAX_NPARTITIONS_SN2)
214 #define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
215 #define XPC_DISENGAGE_REQUEST_AMO (XPC_ENGAGED_PARTITIONS_AMO + 1)
218 * The following structure describes the per partition specific variables.
220 * An array of these structures, one per partition, will be defined. As a
221 * partition becomes active XPC will copy the array entry corresponding to
222 * itself from that partition. It is desirable that the size of this
223 * structure evenly divide into a cacheline, such that none of the entries
224 * in this array crosses a cacheline boundary. As it is now, each entry
225 * occupies half a cacheline.
227 struct xpc_vars_part {
230 u64 openclose_args_pa; /* physical address of open and close args */
231 u64 GPs_pa; /* physical address of Get/Put values */
233 u64 IPI_amo_pa; /* physical address of IPI AMO_t structure */
234 int IPI_nasid; /* nasid of where to send IPIs */
235 int IPI_phys_cpuid; /* physical CPU ID of where to send IPIs */
237 u8 nchannels; /* #of defined channels supported */
239 u8 reserved[23]; /* pad to a full 64 bytes */
243 * The vars_part MAGIC numbers play a part in the first contact protocol.
245 * MAGIC1 indicates that the per partition specific variables for a remote
246 * partition have been initialized by this partition.
248 * MAGIC2 indicates that this partition has pulled the remote partititions
249 * per partition variables that pertain to this partition.
251 #define XPC_VP_MAGIC1 0x0053524156435058L /* 'XPCVARS\0'L (little endian) */
252 #define XPC_VP_MAGIC2 0x0073726176435058L /* 'XPCvars\0'L (little endian) */
254 /* the reserved page sizes and offsets */
256 #define XPC_RP_HEADER_SIZE L1_CACHE_ALIGN(sizeof(struct xpc_rsvd_page))
257 #define XPC_RP_VARS_SIZE L1_CACHE_ALIGN(sizeof(struct xpc_vars))
259 #define XPC_RP_PART_NASIDS(_rp) ((u64 *)((u8 *)(_rp) + XPC_RP_HEADER_SIZE))
260 #define XPC_RP_MACH_NASIDS(_rp) (XPC_RP_PART_NASIDS(_rp) + xp_nasid_mask_words)
261 #define XPC_RP_VARS(_rp) ((struct xpc_vars *)(XPC_RP_MACH_NASIDS(_rp) + \
262 xp_nasid_mask_words))
263 #define XPC_RP_VARS_PART(_rp) ((struct xpc_vars_part *) \
264 ((u8 *)XPC_RP_VARS(_rp) + XPC_RP_VARS_SIZE))
267 * Functions registered by add_timer() or called by kernel_thread() only
268 * allow for a single 64-bit argument. The following macros can be used to
269 * pack and unpack two (32-bit, 16-bit or 8-bit) arguments into or out from
270 * the passed argument.
272 #define XPC_PACK_ARGS(_arg1, _arg2) \
273 ((((u64) _arg1) & 0xffffffff) | \
274 ((((u64) _arg2) & 0xffffffff) << 32))
276 #define XPC_UNPACK_ARG1(_args) (((u64) _args) & 0xffffffff)
277 #define XPC_UNPACK_ARG2(_args) ((((u64) _args) >> 32) & 0xffffffff)
280 * Define a Get/Put value pair (pointers) used with a message queue.
283 s64 get; /* Get value */
284 s64 put; /* Put value */
287 #define XPC_GP_SIZE \
288 L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_MAX_NCHANNELS)
291 * Define a structure that contains arguments associated with opening and
294 struct xpc_openclose_args {
295 u16 reason; /* reason why channel is closing */
296 u16 msg_size; /* sizeof each message entry */
297 u16 remote_nentries; /* #of message entries in remote msg queue */
298 u16 local_nentries; /* #of message entries in local msg queue */
299 u64 local_msgqueue_pa; /* physical address of local message queue */
302 #define XPC_OPENCLOSE_ARGS_SIZE \
303 L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * \
306 /* struct xpc_msg flags */
308 #define XPC_M_DONE 0x01 /* msg has been received/consumed */
309 #define XPC_M_READY 0x02 /* msg is ready to be sent */
310 #define XPC_M_INTERRUPT 0x04 /* send interrupt when msg consumed */
312 #define XPC_MSG_ADDRESS(_payload) \
313 ((struct xpc_msg *)((u8 *)(_payload) - XPC_MSG_PAYLOAD_OFFSET))
316 * Defines notify entry.
318 * This is used to notify a message's sender that their message was received
319 * and consumed by the intended recipient.
322 u8 type; /* type of notification */
324 /* the following two fields are only used if type == XPC_N_CALL */
325 xpc_notify_func func; /* user's notify function */
326 void *key; /* pointer to user's key */
329 /* struct xpc_notify type of notification */
331 #define XPC_N_CALL 0x01 /* notify function provided by user */
334 * Define the structure that manages all the stuff required by a channel. In
335 * particular, they are used to manage the messages sent across the channel.
337 * This structure is private to a partition, and is NOT shared across the
338 * partition boundary.
340 * There is an array of these structures for each remote partition. It is
341 * allocated at the time a partition becomes active. The array contains one
342 * of these structures for each potential channel connection to that partition.
344 * Each of these structures manages two message queues (circular buffers).
345 * They are allocated at the time a channel connection is made. One of
346 * these message queues (local_msgqueue) holds the locally created messages
347 * that are destined for the remote partition. The other of these message
348 * queues (remote_msgqueue) is a locally cached copy of the remote partition's
349 * own local_msgqueue.
351 * The following is a description of the Get/Put pointers used to manage these
352 * two message queues. Consider the local_msgqueue to be on one partition
353 * and the remote_msgqueue to be its cached copy on another partition. A
354 * description of what each of the lettered areas contains is included.
357 * local_msgqueue remote_msgqueue
359 * |/////////| |/////////|
360 * w_remote_GP.get --> +---------+ |/////////|
362 * remote_GP.get --> +---------+ +---------+ <-- local_GP->get
366 * | | +---------+ <-- w_local_GP.get
370 * | | +---------+ <-- w_remote_GP.put
372 * local_GP->put --> +---------+ +---------+ <-- remote_GP.put
376 * w_local_GP.put --> +---------+ |/////////|
377 * |/////////| |/////////|
380 * ( remote_GP.[get|put] are cached copies of the remote
381 * partition's local_GP->[get|put], and thus their values can
382 * lag behind their counterparts on the remote partition. )
385 * A - Messages that have been allocated, but have not yet been sent to the
388 * B - Messages that have been sent, but have not yet been acknowledged by the
389 * remote partition as having been received.
391 * C - Area that needs to be prepared for the copying of sent messages, by
392 * the clearing of the message flags of any previously received messages.
394 * D - Area into which sent messages are to be copied from the remote
395 * partition's local_msgqueue and then delivered to their intended
396 * recipients. [ To allow for a multi-message copy, another pointer
397 * (next_msg_to_pull) has been added to keep track of the next message
398 * number needing to be copied (pulled). It chases after w_remote_GP.put.
399 * Any messages lying between w_local_GP.get and next_msg_to_pull have
400 * been copied and are ready to be delivered. ]
402 * E - Messages that have been copied and delivered, but have not yet been
403 * acknowledged by the recipient as having been received.
405 * F - Messages that have been acknowledged, but XPC has not yet notified the
406 * sender that the message was received by its intended recipient.
407 * This is also an area that needs to be prepared for the allocating of
408 * new messages, by the clearing of the message flags of the acknowledged
412 short partid; /* ID of remote partition connected */
413 spinlock_t lock; /* lock for updating this structure */
414 u32 flags; /* general flags */
416 enum xp_retval reason; /* reason why channel is disconnect'g */
417 int reason_line; /* line# disconnect initiated from */
419 u16 number; /* channel # */
421 u16 msg_size; /* sizeof each msg entry */
422 u16 local_nentries; /* #of msg entries in local msg queue */
423 u16 remote_nentries; /* #of msg entries in remote msg queue */
425 void *local_msgqueue_base; /* base address of kmalloc'd space */
426 struct xpc_msg *local_msgqueue; /* local message queue */
427 void *remote_msgqueue_base; /* base address of kmalloc'd space */
428 struct xpc_msg *remote_msgqueue; /* cached copy of remote partition's */
429 /* local message queue */
430 u64 remote_msgqueue_pa; /* phys addr of remote partition's */
431 /* local message queue */
433 atomic_t references; /* #of external references to queues */
435 atomic_t n_on_msg_allocate_wq; /* #on msg allocation wait queue */
436 wait_queue_head_t msg_allocate_wq; /* msg allocation wait queue */
438 u8 delayed_IPI_flags; /* IPI flags received, but delayed */
439 /* action until channel disconnected */
441 /* queue of msg senders who want to be notified when msg received */
443 atomic_t n_to_notify; /* #of msg senders to notify */
444 struct xpc_notify *notify_queue; /* notify queue for messages sent */
446 xpc_channel_func func; /* user's channel function */
447 void *key; /* pointer to user's key */
449 struct mutex msg_to_pull_mutex; /* next msg to pull serialization */
450 struct completion wdisconnect_wait; /* wait for channel disconnect */
452 struct xpc_openclose_args *local_openclose_args; /* args passed on */
453 /* opening or closing of channel */
455 /* various flavors of local and remote Get/Put values */
457 struct xpc_gp *local_GP; /* local Get/Put values */
458 struct xpc_gp remote_GP; /* remote Get/Put values */
459 struct xpc_gp w_local_GP; /* working local Get/Put values */
460 struct xpc_gp w_remote_GP; /* working remote Get/Put values */
461 s64 next_msg_to_pull; /* Put value of next msg to pull */
463 /* kthread management related fields */
465 atomic_t kthreads_assigned; /* #of kthreads assigned to channel */
466 u32 kthreads_assigned_limit; /* limit on #of kthreads assigned */
467 atomic_t kthreads_idle; /* #of kthreads idle waiting for work */
468 u32 kthreads_idle_limit; /* limit on #of kthreads idle */
469 atomic_t kthreads_active; /* #of kthreads actively working */
471 wait_queue_head_t idle_wq; /* idle kthread wait queue */
473 } ____cacheline_aligned;
475 /* struct xpc_channel flags */
477 #define XPC_C_WASCONNECTED 0x00000001 /* channel was connected */
479 #define XPC_C_ROPENREPLY 0x00000002 /* remote open channel reply */
480 #define XPC_C_OPENREPLY 0x00000004 /* local open channel reply */
481 #define XPC_C_ROPENREQUEST 0x00000008 /* remote open channel request */
482 #define XPC_C_OPENREQUEST 0x00000010 /* local open channel request */
484 #define XPC_C_SETUP 0x00000020 /* channel's msgqueues are alloc'd */
485 #define XPC_C_CONNECTEDCALLOUT 0x00000040 /* connected callout initiated */
486 #define XPC_C_CONNECTEDCALLOUT_MADE \
487 0x00000080 /* connected callout completed */
488 #define XPC_C_CONNECTED 0x00000100 /* local channel is connected */
489 #define XPC_C_CONNECTING 0x00000200 /* channel is being connected */
491 #define XPC_C_RCLOSEREPLY 0x00000400 /* remote close channel reply */
492 #define XPC_C_CLOSEREPLY 0x00000800 /* local close channel reply */
493 #define XPC_C_RCLOSEREQUEST 0x00001000 /* remote close channel request */
494 #define XPC_C_CLOSEREQUEST 0x00002000 /* local close channel request */
496 #define XPC_C_DISCONNECTED 0x00004000 /* channel is disconnected */
497 #define XPC_C_DISCONNECTING 0x00008000 /* channel is being disconnected */
498 #define XPC_C_DISCONNECTINGCALLOUT \
499 0x00010000 /* disconnecting callout initiated */
500 #define XPC_C_DISCONNECTINGCALLOUT_MADE \
501 0x00020000 /* disconnecting callout completed */
502 #define XPC_C_WDISCONNECT 0x00040000 /* waiting for channel disconnect */
505 * Manages channels on a partition basis. There is one of these structures
506 * for each partition (a partition will never utilize the structure that
507 * represents itself).
509 struct xpc_partition {
511 /* XPC HB infrastructure */
513 u8 remote_rp_version; /* version# of partition's rsvd pg */
514 struct timespec remote_rp_stamp; /* time when rsvd pg was initialized */
515 u64 remote_rp_pa; /* phys addr of partition's rsvd pg */
516 u64 remote_vars_pa; /* phys addr of partition's vars */
517 u64 remote_vars_part_pa; /* phys addr of partition's vars part */
518 u64 last_heartbeat; /* HB at last read */
519 u64 remote_amos_page_pa; /* phys addr of partition's amos page */
520 int remote_act_nasid; /* active part's act/deact nasid */
521 int remote_act_phys_cpuid; /* active part's act/deact phys cpuid */
522 u32 act_IRQ_rcvd; /* IRQs since activation */
523 spinlock_t act_lock; /* protect updating of act_state */
524 u8 act_state; /* from XPC HB viewpoint */
525 u8 remote_vars_version; /* version# of partition's vars */
526 enum xp_retval reason; /* reason partition is deactivating */
527 int reason_line; /* line# deactivation initiated from */
528 int reactivate_nasid; /* nasid in partition to reactivate */
530 unsigned long disengage_request_timeout; /* timeout in jiffies */
531 struct timer_list disengage_request_timer;
533 /* XPC infrastructure referencing and teardown control */
535 u8 setup_state; /* infrastructure setup state */
536 wait_queue_head_t teardown_wq; /* kthread waiting to teardown infra */
537 atomic_t references; /* #of references to infrastructure */
540 * NONE OF THE PRECEDING FIELDS OF THIS STRUCTURE WILL BE CLEARED WHEN
541 * XPC SETS UP THE NECESSARY INFRASTRUCTURE TO SUPPORT CROSS PARTITION
542 * COMMUNICATION. ALL OF THE FOLLOWING FIELDS WILL BE CLEARED. (THE
543 * 'nchannels' FIELD MUST BE THE FIRST OF THE FIELDS TO BE CLEARED.)
546 u8 nchannels; /* #of defined channels supported */
547 atomic_t nchannels_active; /* #of channels that are not DISCONNECTED */
548 atomic_t nchannels_engaged; /* #of channels engaged with remote part */
549 struct xpc_channel *channels; /* array of channel structures */
551 void *local_GPs_base; /* base address of kmalloc'd space */
552 struct xpc_gp *local_GPs; /* local Get/Put values */
553 void *remote_GPs_base; /* base address of kmalloc'd space */
554 struct xpc_gp *remote_GPs; /* copy of remote partition's local */
556 u64 remote_GPs_pa; /* phys address of remote partition's local */
559 /* fields used to pass args when opening or closing a channel */
561 void *local_openclose_args_base; /* base address of kmalloc'd space */
562 struct xpc_openclose_args *local_openclose_args; /* local's args */
563 void *remote_openclose_args_base; /* base address of kmalloc'd space */
564 struct xpc_openclose_args *remote_openclose_args; /* copy of remote's */
566 u64 remote_openclose_args_pa; /* phys addr of remote's args */
568 /* IPI sending, receiving and handling related fields */
570 int remote_IPI_nasid; /* nasid of where to send IPIs */
571 int remote_IPI_phys_cpuid; /* phys CPU ID of where to send IPIs */
572 AMO_t *remote_IPI_amo_va; /* address of remote IPI AMO_t structure */
574 AMO_t *local_IPI_amo_va; /* address of IPI AMO_t structure */
575 u64 local_IPI_amo; /* IPI amo flags yet to be handled */
576 char IPI_owner[8]; /* IPI owner's name */
577 struct timer_list dropped_IPI_timer; /* dropped IPI timer */
579 spinlock_t IPI_lock; /* IPI handler lock */
581 /* channel manager related fields */
583 atomic_t channel_mgr_requests; /* #of requests to activate chan mgr */
584 wait_queue_head_t channel_mgr_wq; /* channel mgr's wait queue */
586 } ____cacheline_aligned;
588 /* struct xpc_partition act_state values (for XPC HB) */
590 #define XPC_P_INACTIVE 0x00 /* partition is not active */
591 #define XPC_P_ACTIVATION_REQ 0x01 /* created thread to activate */
592 #define XPC_P_ACTIVATING 0x02 /* activation thread started */
593 #define XPC_P_ACTIVE 0x03 /* xpc_partition_up() was called */
594 #define XPC_P_DEACTIVATING 0x04 /* partition deactivation initiated */
596 #define XPC_DEACTIVATE_PARTITION(_p, _reason) \
597 xpc_deactivate_partition(__LINE__, (_p), (_reason))
599 /* struct xpc_partition setup_state values */
601 #define XPC_P_UNSET 0x00 /* infrastructure was never setup */
602 #define XPC_P_SETUP 0x01 /* infrastructure is setup */
603 #define XPC_P_WTEARDOWN 0x02 /* waiting to teardown infrastructure */
604 #define XPC_P_TORNDOWN 0x03 /* infrastructure is torndown */
607 * struct xpc_partition IPI_timer #of seconds to wait before checking for
608 * dropped IPIs. These occur whenever an IPI amo write doesn't complete until
609 * after the IPI was received.
611 #define XPC_P_DROPPED_IPI_WAIT (0.25 * HZ)
613 /* number of seconds to wait for other partitions to disengage */
614 #define XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT 90
616 /* interval in seconds to print 'waiting disengagement' messages */
617 #define XPC_DISENGAGE_PRINTMSG_INTERVAL 10
619 #define XPC_PARTID(_p) ((short)((_p) - &xpc_partitions[0]))
621 /* found in xp_main.c */
622 extern struct xpc_registration xpc_registrations[];
624 /* found in xpc_main.c */
625 extern struct device *xpc_part;
626 extern struct device *xpc_chan;
627 extern int xpc_disengage_request_timelimit;
628 extern int xpc_disengage_request_timedout;
629 extern irqreturn_t xpc_notify_IRQ_handler(int, void *);
630 extern void xpc_dropped_IPI_check(struct xpc_partition *);
631 extern void xpc_activate_partition(struct xpc_partition *);
632 extern void xpc_activate_kthreads(struct xpc_channel *, int);
633 extern void xpc_create_kthreads(struct xpc_channel *, int, int);
634 extern void xpc_disconnect_wait(int);
636 /* found in xpc_partition.c */
637 extern int xpc_exiting;
638 extern struct xpc_vars *xpc_vars;
639 extern struct xpc_rsvd_page *xpc_rsvd_page;
640 extern struct xpc_vars_part *xpc_vars_part;
641 extern struct xpc_partition *xpc_partitions;
642 extern char *xpc_remote_copy_buffer;
643 extern void *xpc_remote_copy_buffer_base;
644 extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **);
645 extern struct xpc_rsvd_page *xpc_rsvd_page_init(void);
646 extern void xpc_allow_IPI_ops(void);
647 extern void xpc_restrict_IPI_ops(void);
648 extern int xpc_identify_act_IRQ_sender(void);
649 extern int xpc_partition_disengaged(struct xpc_partition *);
650 extern enum xp_retval xpc_mark_partition_active(struct xpc_partition *);
651 extern void xpc_mark_partition_inactive(struct xpc_partition *);
652 extern void xpc_discovery(void);
653 extern void xpc_check_remote_hb(void);
654 extern void xpc_deactivate_partition(const int, struct xpc_partition *,
656 extern enum xp_retval xpc_initiate_partid_to_nasids(short, void *);
658 /* found in xpc_channel.c */
659 extern void xpc_initiate_connect(int);
660 extern void xpc_initiate_disconnect(int);
661 extern enum xp_retval xpc_initiate_allocate(short, int, u32, void **);
662 extern enum xp_retval xpc_initiate_send(short, int, void *);
663 extern enum xp_retval xpc_initiate_send_notify(short, int, void *,
664 xpc_notify_func, void *);
665 extern void xpc_initiate_received(short, int, void *);
666 extern enum xp_retval xpc_setup_infrastructure(struct xpc_partition *);
667 extern enum xp_retval xpc_pull_remote_vars_part(struct xpc_partition *);
668 extern void xpc_process_channel_activity(struct xpc_partition *);
669 extern void xpc_connected_callout(struct xpc_channel *);
670 extern void xpc_deliver_msg(struct xpc_channel *);
671 extern void xpc_disconnect_channel(const int, struct xpc_channel *,
672 enum xp_retval, unsigned long *);
673 extern void xpc_disconnect_callout(struct xpc_channel *, enum xp_retval);
674 extern void xpc_partition_going_down(struct xpc_partition *, enum xp_retval);
675 extern void xpc_teardown_infrastructure(struct xpc_partition *);
678 xpc_wakeup_channel_mgr(struct xpc_partition *part)
680 if (atomic_inc_return(&part->channel_mgr_requests) == 1)
681 wake_up(&part->channel_mgr_wq);
685 * These next two inlines are used to keep us from tearing down a channel's
686 * msg queues while a thread may be referencing them.
689 xpc_msgqueue_ref(struct xpc_channel *ch)
691 atomic_inc(&ch->references);
695 xpc_msgqueue_deref(struct xpc_channel *ch)
697 s32 refs = atomic_dec_return(&ch->references);
701 xpc_wakeup_channel_mgr(&xpc_partitions[ch->partid]);
704 #define XPC_DISCONNECT_CHANNEL(_ch, _reason, _irqflgs) \
705 xpc_disconnect_channel(__LINE__, _ch, _reason, _irqflgs)
708 * These two inlines are used to keep us from tearing down a partition's
709 * setup infrastructure while a thread may be referencing it.
712 xpc_part_deref(struct xpc_partition *part)
714 s32 refs = atomic_dec_return(&part->references);
717 if (refs == 0 && part->setup_state == XPC_P_WTEARDOWN)
718 wake_up(&part->teardown_wq);
722 xpc_part_ref(struct xpc_partition *part)
726 atomic_inc(&part->references);
727 setup = (part->setup_state == XPC_P_SETUP);
729 xpc_part_deref(part);
735 * The following macro is to be used for the setting of the reason and
736 * reason_line fields in both the struct xpc_channel and struct xpc_partition
739 #define XPC_SET_REASON(_p, _reason, _line) \
741 (_p)->reason = _reason; \
742 (_p)->reason_line = _line; \
746 * This next set of inlines are used to keep track of when a partition is
747 * potentially engaged in accessing memory belonging to another partition.
751 xpc_mark_partition_engaged(struct xpc_partition *part)
753 unsigned long irq_flags;
754 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
755 (XPC_ENGAGED_PARTITIONS_AMO *
758 local_irq_save(irq_flags);
760 /* set bit corresponding to our partid in remote partition's AMO */
761 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
762 (1UL << sn_partition_id));
764 * We must always use the nofault function regardless of whether we
765 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
766 * didn't, we'd never know that the other partition is down and would
767 * keep sending IPIs and AMOs to it until the heartbeat times out.
769 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
771 xp_nofault_PIOR_target));
773 local_irq_restore(irq_flags);
777 xpc_mark_partition_disengaged(struct xpc_partition *part)
779 unsigned long irq_flags;
780 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
781 (XPC_ENGAGED_PARTITIONS_AMO *
784 local_irq_save(irq_flags);
786 /* clear bit corresponding to our partid in remote partition's AMO */
787 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
788 ~(1UL << sn_partition_id));
790 * We must always use the nofault function regardless of whether we
791 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
792 * didn't, we'd never know that the other partition is down and would
793 * keep sending IPIs and AMOs to it until the heartbeat times out.
795 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
797 xp_nofault_PIOR_target));
799 local_irq_restore(irq_flags);
803 xpc_request_partition_disengage(struct xpc_partition *part)
805 unsigned long irq_flags;
806 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
807 (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
809 local_irq_save(irq_flags);
811 /* set bit corresponding to our partid in remote partition's AMO */
812 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR,
813 (1UL << sn_partition_id));
815 * We must always use the nofault function regardless of whether we
816 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
817 * didn't, we'd never know that the other partition is down and would
818 * keep sending IPIs and AMOs to it until the heartbeat times out.
820 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
822 xp_nofault_PIOR_target));
824 local_irq_restore(irq_flags);
828 xpc_cancel_partition_disengage_request(struct xpc_partition *part)
830 unsigned long irq_flags;
831 AMO_t *amo = (AMO_t *)__va(part->remote_amos_page_pa +
832 (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
834 local_irq_save(irq_flags);
836 /* clear bit corresponding to our partid in remote partition's AMO */
837 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
838 ~(1UL << sn_partition_id));
840 * We must always use the nofault function regardless of whether we
841 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
842 * didn't, we'd never know that the other partition is down and would
843 * keep sending IPIs and AMOs to it until the heartbeat times out.
845 (void)xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->
847 xp_nofault_PIOR_target));
849 local_irq_restore(irq_flags);
853 xpc_partition_engaged(u64 partid_mask)
855 AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
857 /* return our partition's AMO variable ANDed with partid_mask */
858 return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
863 xpc_partition_disengage_requested(u64 partid_mask)
865 AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
867 /* return our partition's AMO variable ANDed with partid_mask */
868 return (FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_LOAD) &
873 xpc_clear_partition_engaged(u64 partid_mask)
875 AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
877 /* clear bit(s) based on partid_mask in our partition's AMO */
878 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
883 xpc_clear_partition_disengage_request(u64 partid_mask)
885 AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
887 /* clear bit(s) based on partid_mask in our partition's AMO */
888 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_AND,
893 * The following set of macros and inlines are used for the sending and
894 * receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
895 * one that is associated with partition activity (SGI_XPC_ACTIVATE) and
896 * the other that is associated with channel activity (SGI_XPC_NOTIFY).
900 xpc_IPI_receive(AMO_t *amo)
902 return FETCHOP_LOAD_OP(TO_AMO((u64)&amo->variable), FETCHOP_CLEAR);
905 static inline enum xp_retval
906 xpc_IPI_send(AMO_t *amo, u64 flag, int nasid, int phys_cpuid, int vector)
909 unsigned long irq_flags;
911 local_irq_save(irq_flags);
913 FETCHOP_STORE_OP(TO_AMO((u64)&amo->variable), FETCHOP_OR, flag);
914 sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
917 * We must always use the nofault function regardless of whether we
918 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
919 * didn't, we'd never know that the other partition is down and would
920 * keep sending IPIs and AMOs to it until the heartbeat times out.
922 ret = xp_nofault_PIOR((u64 *)GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
923 xp_nofault_PIOR_target));
925 local_irq_restore(irq_flags);
927 return ((ret == 0) ? xpSuccess : xpPioReadError);
931 * IPIs associated with SGI_XPC_ACTIVATE IRQ.
935 * Flag the appropriate AMO variable and send an IPI to the specified node.
938 xpc_activate_IRQ_send(u64 amos_page_pa, int from_nasid, int to_nasid,
941 int w_index = XPC_NASID_W_INDEX(from_nasid);
942 int b_index = XPC_NASID_B_INDEX(from_nasid);
943 AMO_t *amos = (AMO_t *)__va(amos_page_pa +
944 (XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
946 (void)xpc_IPI_send(&amos[w_index], (1UL << b_index), to_nasid,
947 to_phys_cpuid, SGI_XPC_ACTIVATE);
951 xpc_IPI_send_activate(struct xpc_vars *vars)
953 xpc_activate_IRQ_send(vars->amos_page_pa, cnodeid_to_nasid(0),
954 vars->act_nasid, vars->act_phys_cpuid);
958 xpc_IPI_send_activated(struct xpc_partition *part)
960 xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
961 part->remote_act_nasid,
962 part->remote_act_phys_cpuid);
966 xpc_IPI_send_reactivate(struct xpc_partition *part)
968 xpc_activate_IRQ_send(xpc_vars->amos_page_pa, part->reactivate_nasid,
969 xpc_vars->act_nasid, xpc_vars->act_phys_cpuid);
973 xpc_IPI_send_disengage(struct xpc_partition *part)
975 xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
976 part->remote_act_nasid,
977 part->remote_act_phys_cpuid);
981 * IPIs associated with SGI_XPC_NOTIFY IRQ.
985 * Send an IPI to the remote partition that is associated with the
988 #define XPC_NOTIFY_IRQ_SEND(_ch, _ipi_f, _irq_f) \
989 xpc_notify_IRQ_send(_ch, _ipi_f, #_ipi_f, _irq_f)
992 xpc_notify_IRQ_send(struct xpc_channel *ch, u8 ipi_flag, char *ipi_flag_string,
993 unsigned long *irq_flags)
995 struct xpc_partition *part = &xpc_partitions[ch->partid];
998 if (likely(part->act_state != XPC_P_DEACTIVATING)) {
999 ret = xpc_IPI_send(part->remote_IPI_amo_va,
1000 (u64)ipi_flag << (ch->number * 8),
1001 part->remote_IPI_nasid,
1002 part->remote_IPI_phys_cpuid, SGI_XPC_NOTIFY);
1003 dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
1004 ipi_flag_string, ch->partid, ch->number, ret);
1005 if (unlikely(ret != xpSuccess)) {
1006 if (irq_flags != NULL)
1007 spin_unlock_irqrestore(&ch->lock, *irq_flags);
1008 XPC_DEACTIVATE_PARTITION(part, ret);
1009 if (irq_flags != NULL)
1010 spin_lock_irqsave(&ch->lock, *irq_flags);
1016 * Make it look like the remote partition, which is associated with the
1017 * specified channel, sent us an IPI. This faked IPI will be handled
1018 * by xpc_dropped_IPI_check().
1020 #define XPC_NOTIFY_IRQ_SEND_LOCAL(_ch, _ipi_f) \
1021 xpc_notify_IRQ_send_local(_ch, _ipi_f, #_ipi_f)
1024 xpc_notify_IRQ_send_local(struct xpc_channel *ch, u8 ipi_flag,
1025 char *ipi_flag_string)
1027 struct xpc_partition *part = &xpc_partitions[ch->partid];
1029 FETCHOP_STORE_OP(TO_AMO((u64)&part->local_IPI_amo_va->variable),
1030 FETCHOP_OR, ((u64)ipi_flag << (ch->number * 8)));
1031 dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
1032 ipi_flag_string, ch->partid, ch->number);
1036 * The sending and receiving of IPIs includes the setting of an AMO variable
1037 * to indicate the reason the IPI was sent. The 64-bit variable is divided
1038 * up into eight bytes, ordered from right to left. Byte zero pertains to
1039 * channel 0, byte one to channel 1, and so on. Each byte is described by
1040 * the following IPI flags.
1043 #define XPC_IPI_CLOSEREQUEST 0x01
1044 #define XPC_IPI_CLOSEREPLY 0x02
1045 #define XPC_IPI_OPENREQUEST 0x04
1046 #define XPC_IPI_OPENREPLY 0x08
1047 #define XPC_IPI_MSGREQUEST 0x10
1049 /* given an AMO variable and a channel#, get its associated IPI flags */
1050 #define XPC_GET_IPI_FLAGS(_amo, _c) ((u8) (((_amo) >> ((_c) * 8)) & 0xff))
1051 #define XPC_SET_IPI_FLAGS(_amo, _c, _f) (_amo) |= ((u64) (_f) << ((_c) * 8))
1053 #define XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(_amo) ((_amo) & 0x0f0f0f0f0f0f0f0fUL)
1054 #define XPC_ANY_MSG_IPI_FLAGS_SET(_amo) ((_amo) & 0x1010101010101010UL)
1057 xpc_IPI_send_closerequest(struct xpc_channel *ch, unsigned long *irq_flags)
1059 struct xpc_openclose_args *args = ch->local_openclose_args;
1061 args->reason = ch->reason;
1063 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREQUEST, irq_flags);
1067 xpc_IPI_send_closereply(struct xpc_channel *ch, unsigned long *irq_flags)
1069 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREPLY, irq_flags);
1073 xpc_IPI_send_openrequest(struct xpc_channel *ch, unsigned long *irq_flags)
1075 struct xpc_openclose_args *args = ch->local_openclose_args;
1077 args->msg_size = ch->msg_size;
1078 args->local_nentries = ch->local_nentries;
1080 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREQUEST, irq_flags);
1084 xpc_IPI_send_openreply(struct xpc_channel *ch, unsigned long *irq_flags)
1086 struct xpc_openclose_args *args = ch->local_openclose_args;
1088 args->remote_nentries = ch->remote_nentries;
1089 args->local_nentries = ch->local_nentries;
1090 args->local_msgqueue_pa = __pa(ch->local_msgqueue);
1092 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREPLY, irq_flags);
1096 xpc_IPI_send_msgrequest(struct xpc_channel *ch)
1098 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_MSGREQUEST, NULL);
1102 xpc_IPI_send_local_msgrequest(struct xpc_channel *ch)
1104 XPC_NOTIFY_IRQ_SEND_LOCAL(ch, XPC_IPI_MSGREQUEST);
1108 >>> this block comment needs to be moved and re-written.
1109 * Memory for XPC's AMO variables is allocated by the MSPEC driver. These
1110 * pages are located in the lowest granule. The lowest granule uses 4k pages
1111 * for cached references and an alternate TLB handler to never provide a
1112 * cacheable mapping for the entire region. This will prevent speculative
1113 * reading of cached copies of our lines from being issued which will cause
1114 * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
1115 * AMO variables (based on xp_max_npartitions) for message notification and an
1116 * additional 128 AMO variables (based on XP_NASID_MASK_WORDS) for partition
1117 * activation and 2 AMO variables for partition deactivation.
1119 static inline AMO_t *
1120 xpc_IPI_init(int index)
1122 AMO_t *amo = xpc_vars->amos_page + index;
1124 (void)xpc_IPI_receive(amo); /* clear AMO variable */
1128 static inline enum xp_retval
1129 xpc_map_bte_errors(bte_result_t error)
1131 return ((error == BTE_SUCCESS) ? xpSuccess : xpBteCopyError);
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 */