aboutsummaryrefslogtreecommitdiffstats
path: root/include
diff options
context:
space:
mode:
Diffstat (limited to 'include')
-rw-r--r--include/asm-ia64/kprobes.h6
-rw-r--r--include/asm-ia64/sn/sn_sal.h5
-rw-r--r--include/asm-ia64/sn/xp.h4
-rw-r--r--include/asm-ia64/sn/xpc.h1274
-rw-r--r--include/asm-ia64/thread_info.h4
5 files changed, 1289 insertions, 4 deletions
diff --git a/include/asm-ia64/kprobes.h b/include/asm-ia64/kprobes.h
index a74b68104559..8c0fc227f0fb 100644
--- a/include/asm-ia64/kprobes.h
+++ b/include/asm-ia64/kprobes.h
@@ -68,10 +68,14 @@ struct prev_kprobe {
68 unsigned long status; 68 unsigned long status;
69}; 69};
70 70
71#define MAX_PARAM_RSE_SIZE (0x60+0x60/0x3f)
71/* per-cpu kprobe control block */ 72/* per-cpu kprobe control block */
72struct kprobe_ctlblk { 73struct kprobe_ctlblk {
73 unsigned long kprobe_status; 74 unsigned long kprobe_status;
74 struct pt_regs jprobe_saved_regs; 75 struct pt_regs jprobe_saved_regs;
76 unsigned long jprobes_saved_stacked_regs[MAX_PARAM_RSE_SIZE];
77 unsigned long *bsp;
78 unsigned long cfm;
75 struct prev_kprobe prev_kprobe; 79 struct prev_kprobe prev_kprobe;
76}; 80};
77 81
@@ -118,5 +122,7 @@ extern int kprobe_exceptions_notify(struct notifier_block *self,
118static inline void jprobe_return(void) 122static inline void jprobe_return(void)
119{ 123{
120} 124}
125extern void invalidate_stacked_regs(void);
126extern void flush_register_stack(void);
121 127
122#endif /* _ASM_KPROBES_H */ 128#endif /* _ASM_KPROBES_H */
diff --git a/include/asm-ia64/sn/sn_sal.h b/include/asm-ia64/sn/sn_sal.h
index 2a8b0d92a5d6..8b9e10e7cdba 100644
--- a/include/asm-ia64/sn/sn_sal.h
+++ b/include/asm-ia64/sn/sn_sal.h
@@ -75,7 +75,8 @@
75#define SN_SAL_IOIF_GET_HUBDEV_INFO 0x02000055 75#define SN_SAL_IOIF_GET_HUBDEV_INFO 0x02000055
76#define SN_SAL_IOIF_GET_PCIBUS_INFO 0x02000056 76#define SN_SAL_IOIF_GET_PCIBUS_INFO 0x02000056
77#define SN_SAL_IOIF_GET_PCIDEV_INFO 0x02000057 77#define SN_SAL_IOIF_GET_PCIDEV_INFO 0x02000057
78#define SN_SAL_IOIF_GET_WIDGET_DMAFLUSH_LIST 0x02000058 78#define SN_SAL_IOIF_GET_WIDGET_DMAFLUSH_LIST 0x02000058 // deprecated
79#define SN_SAL_IOIF_GET_DEVICE_DMAFLUSH_LIST 0x0200005a
79 80
80#define SN_SAL_HUB_ERROR_INTERRUPT 0x02000060 81#define SN_SAL_HUB_ERROR_INTERRUPT 0x02000060
81#define SN_SAL_BTE_RECOVER 0x02000061 82#define SN_SAL_BTE_RECOVER 0x02000061
@@ -1100,7 +1101,7 @@ ia64_sn_bte_recovery(nasid_t nasid)
1100 struct ia64_sal_retval rv; 1101 struct ia64_sal_retval rv;
1101 1102
1102 rv.status = 0; 1103 rv.status = 0;
1103 SAL_CALL_NOLOCK(rv, SN_SAL_BTE_RECOVER, 0, 0, 0, 0, 0, 0, 0); 1104 SAL_CALL_NOLOCK(rv, SN_SAL_BTE_RECOVER, (u64)nasid, 0, 0, 0, 0, 0, 0);
1104 if (rv.status == SALRET_NOT_IMPLEMENTED) 1105 if (rv.status == SALRET_NOT_IMPLEMENTED)
1105 return 0; 1106 return 0;
1106 return (int) rv.status; 1107 return (int) rv.status;
diff --git a/include/asm-ia64/sn/xp.h b/include/asm-ia64/sn/xp.h
index 49faf8f26430..203945ae034e 100644
--- a/include/asm-ia64/sn/xp.h
+++ b/include/asm-ia64/sn/xp.h
@@ -227,7 +227,9 @@ enum xpc_retval {
227 227
228 xpcOpenCloseError, /* 50: channel open/close protocol error */ 228 xpcOpenCloseError, /* 50: channel open/close protocol error */
229 229
230 xpcUnknownReason /* 51: unknown reason -- must be last in list */ 230 xpcDisconnected, /* 51: channel disconnected (closed) */
231
232 xpcUnknownReason /* 52: unknown reason -- must be last in list */
231}; 233};
232 234
233 235
diff --git a/include/asm-ia64/sn/xpc.h b/include/asm-ia64/sn/xpc.h
new file mode 100644
index 000000000000..87e9cd588510
--- /dev/null
+++ b/include/asm-ia64/sn/xpc.h
@@ -0,0 +1,1274 @@
1/*
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
4 * for more details.
5 *
6 * Copyright (c) 2004-2006 Silicon Graphics, Inc. All Rights Reserved.
7 */
8
9
10/*
11 * Cross Partition Communication (XPC) structures and macros.
12 */
13
14#ifndef _ASM_IA64_SN_XPC_H
15#define _ASM_IA64_SN_XPC_H
16
17
18#include <linux/config.h>
19#include <linux/interrupt.h>
20#include <linux/sysctl.h>
21#include <linux/device.h>
22#include <asm/pgtable.h>
23#include <asm/processor.h>
24#include <asm/sn/bte.h>
25#include <asm/sn/clksupport.h>
26#include <asm/sn/addrs.h>
27#include <asm/sn/mspec.h>
28#include <asm/sn/shub_mmr.h>
29#include <asm/sn/xp.h>
30
31
32/*
33 * XPC Version numbers consist of a major and minor number. XPC can always
34 * talk to versions with same major #, and never talk to versions with a
35 * different major #.
36 */
37#define _XPC_VERSION(_maj, _min) (((_maj) << 4) | ((_min) & 0xf))
38#define XPC_VERSION_MAJOR(_v) ((_v) >> 4)
39#define XPC_VERSION_MINOR(_v) ((_v) & 0xf)
40
41
42/*
43 * The next macros define word or bit representations for given
44 * C-brick nasid in either the SAL provided bit array representing
45 * nasids in the partition/machine or the AMO_t array used for
46 * inter-partition initiation communications.
47 *
48 * For SN2 machines, C-Bricks are alway even numbered NASIDs. As
49 * such, some space will be saved by insisting that nasid information
50 * passed from SAL always be packed for C-Bricks and the
51 * cross-partition interrupts use the same packing scheme.
52 */
53#define XPC_NASID_W_INDEX(_n) (((_n) / 64) / 2)
54#define XPC_NASID_B_INDEX(_n) (((_n) / 2) & (64 - 1))
55#define XPC_NASID_IN_ARRAY(_n, _p) ((_p)[XPC_NASID_W_INDEX(_n)] & \
56 (1UL << XPC_NASID_B_INDEX(_n)))
57#define XPC_NASID_FROM_W_B(_w, _b) (((_w) * 64 + (_b)) * 2)
58
59#define XPC_HB_DEFAULT_INTERVAL 5 /* incr HB every x secs */
60#define XPC_HB_CHECK_DEFAULT_INTERVAL 20 /* check HB every x secs */
61
62/* define the process name of HB checker and the CPU it is pinned to */
63#define XPC_HB_CHECK_THREAD_NAME "xpc_hb"
64#define XPC_HB_CHECK_CPU 0
65
66/* define the process name of the discovery thread */
67#define XPC_DISCOVERY_THREAD_NAME "xpc_discovery"
68
69
70/*
71 * the reserved page
72 *
73 * SAL reserves one page of memory per partition for XPC. Though a full page
74 * in length (16384 bytes), its starting address is not page aligned, but it
75 * is cacheline aligned. The reserved page consists of the following:
76 *
77 * reserved page header
78 *
79 * The first cacheline of the reserved page contains the header
80 * (struct xpc_rsvd_page). Before SAL initialization has completed,
81 * SAL has set up the following fields of the reserved page header:
82 * SAL_signature, SAL_version, partid, and nasids_size. The other
83 * fields are set up by XPC. (xpc_rsvd_page points to the local
84 * partition's reserved page.)
85 *
86 * part_nasids mask
87 * mach_nasids mask
88 *
89 * SAL also sets up two bitmaps (or masks), one that reflects the actual
90 * nasids in this partition (part_nasids), and the other that reflects
91 * the actual nasids in the entire machine (mach_nasids). We're only
92 * interested in the even numbered nasids (which contain the processors
93 * and/or memory), so we only need half as many bits to represent the
94 * nasids. The part_nasids mask is located starting at the first cacheline
95 * following the reserved page header. The mach_nasids mask follows right
96 * after the part_nasids mask. The size in bytes of each mask is reflected
97 * by the reserved page header field 'nasids_size'. (Local partition's
98 * mask pointers are xpc_part_nasids and xpc_mach_nasids.)
99 *
100 * vars
101 * vars part
102 *
103 * Immediately following the mach_nasids mask are the XPC variables
104 * required by other partitions. First are those that are generic to all
105 * partitions (vars), followed on the next available cacheline by those
106 * which are partition specific (vars part). These are setup by XPC.
107 * (Local partition's vars pointers are xpc_vars and xpc_vars_part.)
108 *
109 * Note: Until vars_pa is set, the partition XPC code has not been initialized.
110 */
111struct xpc_rsvd_page {
112 u64 SAL_signature; /* SAL: unique signature */
113 u64 SAL_version; /* SAL: version */
114 u8 partid; /* SAL: partition ID */
115 u8 version;
116 u8 pad1[6]; /* align to next u64 in cacheline */
117 volatile u64 vars_pa;
118 struct timespec stamp; /* time when reserved page was setup by XPC */
119 u64 pad2[9]; /* align to last u64 in cacheline */
120 u64 nasids_size; /* SAL: size of each nasid mask in bytes */
121};
122
123#define XPC_RP_VERSION _XPC_VERSION(1,1) /* version 1.1 of the reserved page */
124
125#define XPC_SUPPORTS_RP_STAMP(_version) \
126 (_version >= _XPC_VERSION(1,1))
127
128/*
129 * compare stamps - the return value is:
130 *
131 * < 0, if stamp1 < stamp2
132 * = 0, if stamp1 == stamp2
133 * > 0, if stamp1 > stamp2
134 */
135static inline int
136xpc_compare_stamps(struct timespec *stamp1, struct timespec *stamp2)
137{
138 int ret;
139
140
141 if ((ret = stamp1->tv_sec - stamp2->tv_sec) == 0) {
142 ret = stamp1->tv_nsec - stamp2->tv_nsec;
143 }
144 return ret;
145}
146
147
148/*
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)
151 */
152
153/*
154 * The following structure describes the partition generic variables
155 * needed by other partitions in order to properly initialize.
156 *
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.
161 */
162struct xpc_vars {
163 u8 version;
164 u64 heartbeat;
165 u64 heartbeating_to_mask;
166 u64 heartbeat_offline; /* if 0, heartbeat should be changing */
167 int act_nasid;
168 int act_phys_cpuid;
169 u64 vars_part_pa;
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 */
172};
173
174#define XPC_V_VERSION _XPC_VERSION(3,1) /* version 3.1 of the cross vars */
175
176#define XPC_SUPPORTS_DISENGAGE_REQUEST(_version) \
177 (_version >= _XPC_VERSION(3,1))
178
179
180static inline int
181xpc_hb_allowed(partid_t partid, struct xpc_vars *vars)
182{
183 return ((vars->heartbeating_to_mask & (1UL << partid)) != 0);
184}
185
186static inline void
187xpc_allow_hb(partid_t partid, struct xpc_vars *vars)
188{
189 u64 old_mask, new_mask;
190
191 do {
192 old_mask = vars->heartbeating_to_mask;
193 new_mask = (old_mask | (1UL << partid));
194 } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
195 old_mask);
196}
197
198static inline void
199xpc_disallow_hb(partid_t partid, struct xpc_vars *vars)
200{
201 u64 old_mask, new_mask;
202
203 do {
204 old_mask = vars->heartbeating_to_mask;
205 new_mask = (old_mask & ~(1UL << partid));
206 } while (cmpxchg(&vars->heartbeating_to_mask, old_mask, new_mask) !=
207 old_mask);
208}
209
210
211/*
212 * The AMOs page consists of a number of AMO variables which are divided into
213 * four groups, The first two groups are used to identify an IRQ's sender.
214 * These two groups consist of 64 and 128 AMO variables respectively. The last
215 * two groups, consisting of just one AMO variable each, are used to identify
216 * the remote partitions that are currently engaged (from the viewpoint of
217 * the XPC running on the remote partition).
218 */
219#define XPC_NOTIFY_IRQ_AMOS 0
220#define XPC_ACTIVATE_IRQ_AMOS (XPC_NOTIFY_IRQ_AMOS + XP_MAX_PARTITIONS)
221#define XPC_ENGAGED_PARTITIONS_AMO (XPC_ACTIVATE_IRQ_AMOS + XP_NASID_MASK_WORDS)
222#define XPC_DISENGAGE_REQUEST_AMO (XPC_ENGAGED_PARTITIONS_AMO + 1)
223
224
225/*
226 * The following structure describes the per partition specific variables.
227 *
228 * An array of these structures, one per partition, will be defined. As a
229 * partition becomes active XPC will copy the array entry corresponding to
230 * itself from that partition. It is desirable that the size of this
231 * structure evenly divide into a cacheline, such that none of the entries
232 * in this array crosses a cacheline boundary. As it is now, each entry
233 * occupies half a cacheline.
234 */
235struct xpc_vars_part {
236 volatile u64 magic;
237
238 u64 openclose_args_pa; /* physical address of open and close args */
239 u64 GPs_pa; /* physical address of Get/Put values */
240
241 u64 IPI_amo_pa; /* physical address of IPI AMO_t structure */
242 int IPI_nasid; /* nasid of where to send IPIs */
243 int IPI_phys_cpuid; /* physical CPU ID of where to send IPIs */
244
245 u8 nchannels; /* #of defined channels supported */
246
247 u8 reserved[23]; /* pad to a full 64 bytes */
248};
249
250/*
251 * The vars_part MAGIC numbers play a part in the first contact protocol.
252 *
253 * MAGIC1 indicates that the per partition specific variables for a remote
254 * partition have been initialized by this partition.
255 *
256 * MAGIC2 indicates that this partition has pulled the remote partititions
257 * per partition variables that pertain to this partition.
258 */
259#define XPC_VP_MAGIC1 0x0053524156435058L /* 'XPCVARS\0'L (little endian) */
260#define XPC_VP_MAGIC2 0x0073726176435058L /* 'XPCvars\0'L (little endian) */
261
262
263/* the reserved page sizes and offsets */
264
265#define XPC_RP_HEADER_SIZE L1_CACHE_ALIGN(sizeof(struct xpc_rsvd_page))
266#define XPC_RP_VARS_SIZE L1_CACHE_ALIGN(sizeof(struct xpc_vars))
267
268#define XPC_RP_PART_NASIDS(_rp) (u64 *) ((u8 *) _rp + XPC_RP_HEADER_SIZE)
269#define XPC_RP_MACH_NASIDS(_rp) (XPC_RP_PART_NASIDS(_rp) + xp_nasid_mask_words)
270#define XPC_RP_VARS(_rp) ((struct xpc_vars *) XPC_RP_MACH_NASIDS(_rp) + xp_nasid_mask_words)
271#define XPC_RP_VARS_PART(_rp) (struct xpc_vars_part *) ((u8 *) XPC_RP_VARS(rp) + XPC_RP_VARS_SIZE)
272
273
274/*
275 * Functions registered by add_timer() or called by kernel_thread() only
276 * allow for a single 64-bit argument. The following macros can be used to
277 * pack and unpack two (32-bit, 16-bit or 8-bit) arguments into or out from
278 * the passed argument.
279 */
280#define XPC_PACK_ARGS(_arg1, _arg2) \
281 ((((u64) _arg1) & 0xffffffff) | \
282 ((((u64) _arg2) & 0xffffffff) << 32))
283
284#define XPC_UNPACK_ARG1(_args) (((u64) _args) & 0xffffffff)
285#define XPC_UNPACK_ARG2(_args) ((((u64) _args) >> 32) & 0xffffffff)
286
287
288
289/*
290 * Define a Get/Put value pair (pointers) used with a message queue.
291 */
292struct xpc_gp {
293 volatile s64 get; /* Get value */
294 volatile s64 put; /* Put value */
295};
296
297#define XPC_GP_SIZE \
298 L1_CACHE_ALIGN(sizeof(struct xpc_gp) * XPC_NCHANNELS)
299
300
301
302/*
303 * Define a structure that contains arguments associated with opening and
304 * closing a channel.
305 */
306struct xpc_openclose_args {
307 u16 reason; /* reason why channel is closing */
308 u16 msg_size; /* sizeof each message entry */
309 u16 remote_nentries; /* #of message entries in remote msg queue */
310 u16 local_nentries; /* #of message entries in local msg queue */
311 u64 local_msgqueue_pa; /* physical address of local message queue */
312};
313
314#define XPC_OPENCLOSE_ARGS_SIZE \
315 L1_CACHE_ALIGN(sizeof(struct xpc_openclose_args) * XPC_NCHANNELS)
316
317
318
319/* struct xpc_msg flags */
320
321#define XPC_M_DONE 0x01 /* msg has been received/consumed */
322#define XPC_M_READY 0x02 /* msg is ready to be sent */
323#define XPC_M_INTERRUPT 0x04 /* send interrupt when msg consumed */
324
325
326#define XPC_MSG_ADDRESS(_payload) \
327 ((struct xpc_msg *)((u8 *)(_payload) - XPC_MSG_PAYLOAD_OFFSET))
328
329
330
331/*
332 * Defines notify entry.
333 *
334 * This is used to notify a message's sender that their message was received
335 * and consumed by the intended recipient.
336 */
337struct xpc_notify {
338 struct semaphore sema; /* notify semaphore */
339 volatile u8 type; /* type of notification */
340
341 /* the following two fields are only used if type == XPC_N_CALL */
342 xpc_notify_func func; /* user's notify function */
343 void *key; /* pointer to user's key */
344};
345
346/* struct xpc_notify type of notification */
347
348#define XPC_N_CALL 0x01 /* notify function provided by user */
349
350
351
352/*
353 * Define the structure that manages all the stuff required by a channel. In
354 * particular, they are used to manage the messages sent across the channel.
355 *
356 * This structure is private to a partition, and is NOT shared across the
357 * partition boundary.
358 *
359 * There is an array of these structures for each remote partition. It is
360 * allocated at the time a partition becomes active. The array contains one
361 * of these structures for each potential channel connection to that partition.
362 *
363 * Each of these structures manages two message queues (circular buffers).
364 * They are allocated at the time a channel connection is made. One of
365 * these message queues (local_msgqueue) holds the locally created messages
366 * that are destined for the remote partition. The other of these message
367 * queues (remote_msgqueue) is a locally cached copy of the remote partition's
368 * own local_msgqueue.
369 *
370 * The following is a description of the Get/Put pointers used to manage these
371 * two message queues. Consider the local_msgqueue to be on one partition
372 * and the remote_msgqueue to be its cached copy on another partition. A
373 * description of what each of the lettered areas contains is included.
374 *
375 *
376 * local_msgqueue remote_msgqueue
377 *
378 * |/////////| |/////////|
379 * w_remote_GP.get --> +---------+ |/////////|
380 * | F | |/////////|
381 * remote_GP.get --> +---------+ +---------+ <-- local_GP->get
382 * | | | |
383 * | | | E |
384 * | | | |
385 * | | +---------+ <-- w_local_GP.get
386 * | B | |/////////|
387 * | | |////D////|
388 * | | |/////////|
389 * | | +---------+ <-- w_remote_GP.put
390 * | | |////C////|
391 * local_GP->put --> +---------+ +---------+ <-- remote_GP.put
392 * | | |/////////|
393 * | A | |/////////|
394 * | | |/////////|
395 * w_local_GP.put --> +---------+ |/////////|
396 * |/////////| |/////////|
397 *
398 *
399 * ( remote_GP.[get|put] are cached copies of the remote
400 * partition's local_GP->[get|put], and thus their values can
401 * lag behind their counterparts on the remote partition. )
402 *
403 *
404 * A - Messages that have been allocated, but have not yet been sent to the
405 * remote partition.
406 *
407 * B - Messages that have been sent, but have not yet been acknowledged by the
408 * remote partition as having been received.
409 *
410 * C - Area that needs to be prepared for the copying of sent messages, by
411 * the clearing of the message flags of any previously received messages.
412 *
413 * D - Area into which sent messages are to be copied from the remote
414 * partition's local_msgqueue and then delivered to their intended
415 * recipients. [ To allow for a multi-message copy, another pointer
416 * (next_msg_to_pull) has been added to keep track of the next message
417 * number needing to be copied (pulled). It chases after w_remote_GP.put.
418 * Any messages lying between w_local_GP.get and next_msg_to_pull have
419 * been copied and are ready to be delivered. ]
420 *
421 * E - Messages that have been copied and delivered, but have not yet been
422 * acknowledged by the recipient as having been received.
423 *
424 * F - Messages that have been acknowledged, but XPC has not yet notified the
425 * sender that the message was received by its intended recipient.
426 * This is also an area that needs to be prepared for the allocating of
427 * new messages, by the clearing of the message flags of the acknowledged
428 * messages.
429 */
430struct xpc_channel {
431 partid_t partid; /* ID of remote partition connected */
432 spinlock_t lock; /* lock for updating this structure */
433 u32 flags; /* general flags */
434
435 enum xpc_retval reason; /* reason why channel is disconnect'g */
436 int reason_line; /* line# disconnect initiated from */
437
438 u16 number; /* channel # */
439
440 u16 msg_size; /* sizeof each msg entry */
441 u16 local_nentries; /* #of msg entries in local msg queue */
442 u16 remote_nentries; /* #of msg entries in remote msg queue*/
443
444 void *local_msgqueue_base; /* base address of kmalloc'd space */
445 struct xpc_msg *local_msgqueue; /* local message queue */
446 void *remote_msgqueue_base; /* base address of kmalloc'd space */
447 struct xpc_msg *remote_msgqueue;/* cached copy of remote partition's */
448 /* local message queue */
449 u64 remote_msgqueue_pa; /* phys addr of remote partition's */
450 /* local message queue */
451
452 atomic_t references; /* #of external references to queues */
453
454 atomic_t n_on_msg_allocate_wq; /* #on msg allocation wait queue */
455 wait_queue_head_t msg_allocate_wq; /* msg allocation wait queue */
456
457 u8 delayed_IPI_flags; /* IPI flags received, but delayed */
458 /* action until channel disconnected */
459
460 /* queue of msg senders who want to be notified when msg received */
461
462 atomic_t n_to_notify; /* #of msg senders to notify */
463 struct xpc_notify *notify_queue;/* notify queue for messages sent */
464
465 xpc_channel_func func; /* user's channel function */
466 void *key; /* pointer to user's key */
467
468 struct semaphore msg_to_pull_sema; /* next msg to pull serialization */
469 struct semaphore wdisconnect_sema; /* wait for channel disconnect */
470
471 struct xpc_openclose_args *local_openclose_args; /* args passed on */
472 /* opening or closing of channel */
473
474 /* various flavors of local and remote Get/Put values */
475
476 struct xpc_gp *local_GP; /* local Get/Put values */
477 struct xpc_gp remote_GP; /* remote Get/Put values */
478 struct xpc_gp w_local_GP; /* working local Get/Put values */
479 struct xpc_gp w_remote_GP; /* working remote Get/Put values */
480 s64 next_msg_to_pull; /* Put value of next msg to pull */
481
482 /* kthread management related fields */
483
484// >>> rethink having kthreads_assigned_limit and kthreads_idle_limit; perhaps
485// >>> allow the assigned limit be unbounded and let the idle limit be dynamic
486// >>> dependent on activity over the last interval of time
487 atomic_t kthreads_assigned; /* #of kthreads assigned to channel */
488 u32 kthreads_assigned_limit; /* limit on #of kthreads assigned */
489 atomic_t kthreads_idle; /* #of kthreads idle waiting for work */
490 u32 kthreads_idle_limit; /* limit on #of kthreads idle */
491 atomic_t kthreads_active; /* #of kthreads actively working */
492 // >>> following field is temporary
493 u32 kthreads_created; /* total #of kthreads created */
494
495 wait_queue_head_t idle_wq; /* idle kthread wait queue */
496
497} ____cacheline_aligned;
498
499
500/* struct xpc_channel flags */
501
502#define XPC_C_WASCONNECTED 0x00000001 /* channel was connected */
503
504#define XPC_C_ROPENREPLY 0x00000002 /* remote open channel reply */
505#define XPC_C_OPENREPLY 0x00000004 /* local open channel reply */
506#define XPC_C_ROPENREQUEST 0x00000008 /* remote open channel request */
507#define XPC_C_OPENREQUEST 0x00000010 /* local open channel request */
508
509#define XPC_C_SETUP 0x00000020 /* channel's msgqueues are alloc'd */
510#define XPC_C_CONNECTCALLOUT 0x00000040 /* channel connected callout made */
511#define XPC_C_CONNECTED 0x00000080 /* local channel is connected */
512#define XPC_C_CONNECTING 0x00000100 /* channel is being connected */
513
514#define XPC_C_RCLOSEREPLY 0x00000200 /* remote close channel reply */
515#define XPC_C_CLOSEREPLY 0x00000400 /* local close channel reply */
516#define XPC_C_RCLOSEREQUEST 0x00000800 /* remote close channel request */
517#define XPC_C_CLOSEREQUEST 0x00001000 /* local close channel request */
518
519#define XPC_C_DISCONNECTED 0x00002000 /* channel is disconnected */
520#define XPC_C_DISCONNECTING 0x00004000 /* channel is being disconnected */
521#define XPC_C_DISCONNECTCALLOUT 0x00008000 /* chan disconnected callout made */
522#define XPC_C_WDISCONNECT 0x00010000 /* waiting for channel disconnect */
523
524
525
526/*
527 * Manages channels on a partition basis. There is one of these structures
528 * for each partition (a partition will never utilize the structure that
529 * represents itself).
530 */
531struct xpc_partition {
532
533 /* XPC HB infrastructure */
534
535 u8 remote_rp_version; /* version# of partition's rsvd pg */
536 struct timespec remote_rp_stamp;/* time when rsvd pg was initialized */
537 u64 remote_rp_pa; /* phys addr of partition's rsvd pg */
538 u64 remote_vars_pa; /* phys addr of partition's vars */
539 u64 remote_vars_part_pa; /* phys addr of partition's vars part */
540 u64 last_heartbeat; /* HB at last read */
541 u64 remote_amos_page_pa; /* phys addr of partition's amos page */
542 int remote_act_nasid; /* active part's act/deact nasid */
543 int remote_act_phys_cpuid; /* active part's act/deact phys cpuid */
544 u32 act_IRQ_rcvd; /* IRQs since activation */
545 spinlock_t act_lock; /* protect updating of act_state */
546 u8 act_state; /* from XPC HB viewpoint */
547 u8 remote_vars_version; /* version# of partition's vars */
548 enum xpc_retval reason; /* reason partition is deactivating */
549 int reason_line; /* line# deactivation initiated from */
550 int reactivate_nasid; /* nasid in partition to reactivate */
551
552 unsigned long disengage_request_timeout; /* timeout in jiffies */
553 struct timer_list disengage_request_timer;
554
555
556 /* XPC infrastructure referencing and teardown control */
557
558 volatile u8 setup_state; /* infrastructure setup state */
559 wait_queue_head_t teardown_wq; /* kthread waiting to teardown infra */
560 atomic_t references; /* #of references to infrastructure */
561
562
563 /*
564 * NONE OF THE PRECEDING FIELDS OF THIS STRUCTURE WILL BE CLEARED WHEN
565 * XPC SETS UP THE NECESSARY INFRASTRUCTURE TO SUPPORT CROSS PARTITION
566 * COMMUNICATION. ALL OF THE FOLLOWING FIELDS WILL BE CLEARED. (THE
567 * 'nchannels' FIELD MUST BE THE FIRST OF THE FIELDS TO BE CLEARED.)
568 */
569
570
571 u8 nchannels; /* #of defined channels supported */
572 atomic_t nchannels_active; /* #of channels that are not DISCONNECTED */
573 atomic_t nchannels_engaged;/* #of channels engaged with remote part */
574 struct xpc_channel *channels;/* array of channel structures */
575
576 void *local_GPs_base; /* base address of kmalloc'd space */
577 struct xpc_gp *local_GPs; /* local Get/Put values */
578 void *remote_GPs_base; /* base address of kmalloc'd space */
579 struct xpc_gp *remote_GPs;/* copy of remote partition's local Get/Put */
580 /* values */
581 u64 remote_GPs_pa; /* phys address of remote partition's local */
582 /* Get/Put values */
583
584
585 /* fields used to pass args when opening or closing a channel */
586
587 void *local_openclose_args_base; /* base address of kmalloc'd space */
588 struct xpc_openclose_args *local_openclose_args; /* local's args */
589 void *remote_openclose_args_base; /* base address of kmalloc'd space */
590 struct xpc_openclose_args *remote_openclose_args; /* copy of remote's */
591 /* args */
592 u64 remote_openclose_args_pa; /* phys addr of remote's args */
593
594
595 /* IPI sending, receiving and handling related fields */
596
597 int remote_IPI_nasid; /* nasid of where to send IPIs */
598 int remote_IPI_phys_cpuid; /* phys CPU ID of where to send IPIs */
599 AMO_t *remote_IPI_amo_va; /* address of remote IPI AMO_t structure */
600
601 AMO_t *local_IPI_amo_va; /* address of IPI AMO_t structure */
602 u64 local_IPI_amo; /* IPI amo flags yet to be handled */
603 char IPI_owner[8]; /* IPI owner's name */
604 struct timer_list dropped_IPI_timer; /* dropped IPI timer */
605
606 spinlock_t IPI_lock; /* IPI handler lock */
607
608
609 /* channel manager related fields */
610
611 atomic_t channel_mgr_requests; /* #of requests to activate chan mgr */
612 wait_queue_head_t channel_mgr_wq; /* channel mgr's wait queue */
613
614} ____cacheline_aligned;
615
616
617/* struct xpc_partition act_state values (for XPC HB) */
618
619#define XPC_P_INACTIVE 0x00 /* partition is not active */
620#define XPC_P_ACTIVATION_REQ 0x01 /* created thread to activate */
621#define XPC_P_ACTIVATING 0x02 /* activation thread started */
622#define XPC_P_ACTIVE 0x03 /* xpc_partition_up() was called */
623#define XPC_P_DEACTIVATING 0x04 /* partition deactivation initiated */
624
625
626#define XPC_DEACTIVATE_PARTITION(_p, _reason) \
627 xpc_deactivate_partition(__LINE__, (_p), (_reason))
628
629
630/* struct xpc_partition setup_state values */
631
632#define XPC_P_UNSET 0x00 /* infrastructure was never setup */
633#define XPC_P_SETUP 0x01 /* infrastructure is setup */
634#define XPC_P_WTEARDOWN 0x02 /* waiting to teardown infrastructure */
635#define XPC_P_TORNDOWN 0x03 /* infrastructure is torndown */
636
637
638
639/*
640 * struct xpc_partition IPI_timer #of seconds to wait before checking for
641 * dropped IPIs. These occur whenever an IPI amo write doesn't complete until
642 * after the IPI was received.
643 */
644#define XPC_P_DROPPED_IPI_WAIT (0.25 * HZ)
645
646
647/* number of seconds to wait for other partitions to disengage */
648#define XPC_DISENGAGE_REQUEST_DEFAULT_TIMELIMIT 90
649
650/* interval in seconds to print 'waiting disengagement' messages */
651#define XPC_DISENGAGE_PRINTMSG_INTERVAL 10
652
653
654#define XPC_PARTID(_p) ((partid_t) ((_p) - &xpc_partitions[0]))
655
656
657
658/* found in xp_main.c */
659extern struct xpc_registration xpc_registrations[];
660
661
662/* found in xpc_main.c */
663extern struct device *xpc_part;
664extern struct device *xpc_chan;
665extern int xpc_disengage_request_timelimit;
666extern int xpc_disengage_request_timedout;
667extern irqreturn_t xpc_notify_IRQ_handler(int, void *, struct pt_regs *);
668extern void xpc_dropped_IPI_check(struct xpc_partition *);
669extern void xpc_activate_partition(struct xpc_partition *);
670extern void xpc_activate_kthreads(struct xpc_channel *, int);
671extern void xpc_create_kthreads(struct xpc_channel *, int);
672extern void xpc_disconnect_wait(int);
673
674
675/* found in xpc_partition.c */
676extern int xpc_exiting;
677extern struct xpc_vars *xpc_vars;
678extern struct xpc_rsvd_page *xpc_rsvd_page;
679extern struct xpc_vars_part *xpc_vars_part;
680extern struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];
681extern char xpc_remote_copy_buffer[];
682extern struct xpc_rsvd_page *xpc_rsvd_page_init(void);
683extern void xpc_allow_IPI_ops(void);
684extern void xpc_restrict_IPI_ops(void);
685extern int xpc_identify_act_IRQ_sender(void);
686extern int xpc_partition_disengaged(struct xpc_partition *);
687extern enum xpc_retval xpc_mark_partition_active(struct xpc_partition *);
688extern void xpc_mark_partition_inactive(struct xpc_partition *);
689extern void xpc_discovery(void);
690extern void xpc_check_remote_hb(void);
691extern void xpc_deactivate_partition(const int, struct xpc_partition *,
692 enum xpc_retval);
693extern enum xpc_retval xpc_initiate_partid_to_nasids(partid_t, void *);
694
695
696/* found in xpc_channel.c */
697extern void xpc_initiate_connect(int);
698extern void xpc_initiate_disconnect(int);
699extern enum xpc_retval xpc_initiate_allocate(partid_t, int, u32, void **);
700extern enum xpc_retval xpc_initiate_send(partid_t, int, void *);
701extern enum xpc_retval xpc_initiate_send_notify(partid_t, int, void *,
702 xpc_notify_func, void *);
703extern void xpc_initiate_received(partid_t, int, void *);
704extern enum xpc_retval xpc_setup_infrastructure(struct xpc_partition *);
705extern enum xpc_retval xpc_pull_remote_vars_part(struct xpc_partition *);
706extern void xpc_process_channel_activity(struct xpc_partition *);
707extern void xpc_connected_callout(struct xpc_channel *);
708extern void xpc_deliver_msg(struct xpc_channel *);
709extern void xpc_disconnect_channel(const int, struct xpc_channel *,
710 enum xpc_retval, unsigned long *);
711extern void xpc_disconnect_callout(struct xpc_channel *, enum xpc_retval);
712extern void xpc_partition_going_down(struct xpc_partition *, enum xpc_retval);
713extern void xpc_teardown_infrastructure(struct xpc_partition *);
714
715
716
717static inline void
718xpc_wakeup_channel_mgr(struct xpc_partition *part)
719{
720 if (atomic_inc_return(&part->channel_mgr_requests) == 1) {
721 wake_up(&part->channel_mgr_wq);
722 }
723}
724
725
726
727/*
728 * These next two inlines are used to keep us from tearing down a channel's
729 * msg queues while a thread may be referencing them.
730 */
731static inline void
732xpc_msgqueue_ref(struct xpc_channel *ch)
733{
734 atomic_inc(&ch->references);
735}
736
737static inline void
738xpc_msgqueue_deref(struct xpc_channel *ch)
739{
740 s32 refs = atomic_dec_return(&ch->references);
741
742 DBUG_ON(refs < 0);
743 if (refs == 0) {
744 xpc_wakeup_channel_mgr(&xpc_partitions[ch->partid]);
745 }
746}
747
748
749
750#define XPC_DISCONNECT_CHANNEL(_ch, _reason, _irqflgs) \
751 xpc_disconnect_channel(__LINE__, _ch, _reason, _irqflgs)
752
753
754/*
755 * These two inlines are used to keep us from tearing down a partition's
756 * setup infrastructure while a thread may be referencing it.
757 */
758static inline void
759xpc_part_deref(struct xpc_partition *part)
760{
761 s32 refs = atomic_dec_return(&part->references);
762
763
764 DBUG_ON(refs < 0);
765 if (refs == 0 && part->setup_state == XPC_P_WTEARDOWN) {
766 wake_up(&part->teardown_wq);
767 }
768}
769
770static inline int
771xpc_part_ref(struct xpc_partition *part)
772{
773 int setup;
774
775
776 atomic_inc(&part->references);
777 setup = (part->setup_state == XPC_P_SETUP);
778 if (!setup) {
779 xpc_part_deref(part);
780 }
781 return setup;
782}
783
784
785
786/*
787 * The following macro is to be used for the setting of the reason and
788 * reason_line fields in both the struct xpc_channel and struct xpc_partition
789 * structures.
790 */
791#define XPC_SET_REASON(_p, _reason, _line) \
792 { \
793 (_p)->reason = _reason; \
794 (_p)->reason_line = _line; \
795 }
796
797
798
799/*
800 * This next set of inlines are used to keep track of when a partition is
801 * potentially engaged in accessing memory belonging to another partition.
802 */
803
804static inline void
805xpc_mark_partition_engaged(struct xpc_partition *part)
806{
807 unsigned long irq_flags;
808 AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
809 (XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
810
811
812 local_irq_save(irq_flags);
813
814 /* set bit corresponding to our partid in remote partition's AMO */
815 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
816 (1UL << sn_partition_id));
817 /*
818 * We must always use the nofault function regardless of whether we
819 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
820 * didn't, we'd never know that the other partition is down and would
821 * keep sending IPIs and AMOs to it until the heartbeat times out.
822 */
823 (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
824 variable), xp_nofault_PIOR_target));
825
826 local_irq_restore(irq_flags);
827}
828
829static inline void
830xpc_mark_partition_disengaged(struct xpc_partition *part)
831{
832 unsigned long irq_flags;
833 AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
834 (XPC_ENGAGED_PARTITIONS_AMO * sizeof(AMO_t)));
835
836
837 local_irq_save(irq_flags);
838
839 /* clear bit corresponding to our partid in remote partition's AMO */
840 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
841 ~(1UL << sn_partition_id));
842 /*
843 * We must always use the nofault function regardless of whether we
844 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
845 * didn't, we'd never know that the other partition is down and would
846 * keep sending IPIs and AMOs to it until the heartbeat times out.
847 */
848 (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
849 variable), xp_nofault_PIOR_target));
850
851 local_irq_restore(irq_flags);
852}
853
854static inline void
855xpc_request_partition_disengage(struct xpc_partition *part)
856{
857 unsigned long irq_flags;
858 AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
859 (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
860
861
862 local_irq_save(irq_flags);
863
864 /* set bit corresponding to our partid in remote partition's AMO */
865 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR,
866 (1UL << sn_partition_id));
867 /*
868 * We must always use the nofault function regardless of whether we
869 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
870 * didn't, we'd never know that the other partition is down and would
871 * keep sending IPIs and AMOs to it until the heartbeat times out.
872 */
873 (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
874 variable), xp_nofault_PIOR_target));
875
876 local_irq_restore(irq_flags);
877}
878
879static inline void
880xpc_cancel_partition_disengage_request(struct xpc_partition *part)
881{
882 unsigned long irq_flags;
883 AMO_t *amo = (AMO_t *) __va(part->remote_amos_page_pa +
884 (XPC_DISENGAGE_REQUEST_AMO * sizeof(AMO_t)));
885
886
887 local_irq_save(irq_flags);
888
889 /* clear bit corresponding to our partid in remote partition's AMO */
890 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
891 ~(1UL << sn_partition_id));
892 /*
893 * We must always use the nofault function regardless of whether we
894 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
895 * didn't, we'd never know that the other partition is down and would
896 * keep sending IPIs and AMOs to it until the heartbeat times out.
897 */
898 (void) xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->
899 variable), xp_nofault_PIOR_target));
900
901 local_irq_restore(irq_flags);
902}
903
904static inline u64
905xpc_partition_engaged(u64 partid_mask)
906{
907 AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
908
909
910 /* return our partition's AMO variable ANDed with partid_mask */
911 return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
912 partid_mask);
913}
914
915static inline u64
916xpc_partition_disengage_requested(u64 partid_mask)
917{
918 AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
919
920
921 /* return our partition's AMO variable ANDed with partid_mask */
922 return (FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_LOAD) &
923 partid_mask);
924}
925
926static inline void
927xpc_clear_partition_engaged(u64 partid_mask)
928{
929 AMO_t *amo = xpc_vars->amos_page + XPC_ENGAGED_PARTITIONS_AMO;
930
931
932 /* clear bit(s) based on partid_mask in our partition's AMO */
933 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
934 ~partid_mask);
935}
936
937static inline void
938xpc_clear_partition_disengage_request(u64 partid_mask)
939{
940 AMO_t *amo = xpc_vars->amos_page + XPC_DISENGAGE_REQUEST_AMO;
941
942
943 /* clear bit(s) based on partid_mask in our partition's AMO */
944 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_AND,
945 ~partid_mask);
946}
947
948
949
950/*
951 * The following set of macros and inlines are used for the sending and
952 * receiving of IPIs (also known as IRQs). There are two flavors of IPIs,
953 * one that is associated with partition activity (SGI_XPC_ACTIVATE) and
954 * the other that is associated with channel activity (SGI_XPC_NOTIFY).
955 */
956
957static inline u64
958xpc_IPI_receive(AMO_t *amo)
959{
960 return FETCHOP_LOAD_OP(TO_AMO((u64) &amo->variable), FETCHOP_CLEAR);
961}
962
963
964static inline enum xpc_retval
965xpc_IPI_send(AMO_t *amo, u64 flag, int nasid, int phys_cpuid, int vector)
966{
967 int ret = 0;
968 unsigned long irq_flags;
969
970
971 local_irq_save(irq_flags);
972
973 FETCHOP_STORE_OP(TO_AMO((u64) &amo->variable), FETCHOP_OR, flag);
974 sn_send_IPI_phys(nasid, phys_cpuid, vector, 0);
975
976 /*
977 * We must always use the nofault function regardless of whether we
978 * are on a Shub 1.1 system or a Shub 1.2 slice 0xc processor. If we
979 * didn't, we'd never know that the other partition is down and would
980 * keep sending IPIs and AMOs to it until the heartbeat times out.
981 */
982 ret = xp_nofault_PIOR((u64 *) GLOBAL_MMR_ADDR(NASID_GET(&amo->variable),
983 xp_nofault_PIOR_target));
984
985 local_irq_restore(irq_flags);
986
987 return ((ret == 0) ? xpcSuccess : xpcPioReadError);
988}
989
990
991/*
992 * IPIs associated with SGI_XPC_ACTIVATE IRQ.
993 */
994
995/*
996 * Flag the appropriate AMO variable and send an IPI to the specified node.
997 */
998static inline void
999xpc_activate_IRQ_send(u64 amos_page_pa, int from_nasid, int to_nasid,
1000 int to_phys_cpuid)
1001{
1002 int w_index = XPC_NASID_W_INDEX(from_nasid);
1003 int b_index = XPC_NASID_B_INDEX(from_nasid);
1004 AMO_t *amos = (AMO_t *) __va(amos_page_pa +
1005 (XPC_ACTIVATE_IRQ_AMOS * sizeof(AMO_t)));
1006
1007
1008 (void) xpc_IPI_send(&amos[w_index], (1UL << b_index), to_nasid,
1009 to_phys_cpuid, SGI_XPC_ACTIVATE);
1010}
1011
1012static inline void
1013xpc_IPI_send_activate(struct xpc_vars *vars)
1014{
1015 xpc_activate_IRQ_send(vars->amos_page_pa, cnodeid_to_nasid(0),
1016 vars->act_nasid, vars->act_phys_cpuid);
1017}
1018
1019static inline void
1020xpc_IPI_send_activated(struct xpc_partition *part)
1021{
1022 xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
1023 part->remote_act_nasid, part->remote_act_phys_cpuid);
1024}
1025
1026static inline void
1027xpc_IPI_send_reactivate(struct xpc_partition *part)
1028{
1029 xpc_activate_IRQ_send(xpc_vars->amos_page_pa, part->reactivate_nasid,
1030 xpc_vars->act_nasid, xpc_vars->act_phys_cpuid);
1031}
1032
1033static inline void
1034xpc_IPI_send_disengage(struct xpc_partition *part)
1035{
1036 xpc_activate_IRQ_send(part->remote_amos_page_pa, cnodeid_to_nasid(0),
1037 part->remote_act_nasid, part->remote_act_phys_cpuid);
1038}
1039
1040
1041/*
1042 * IPIs associated with SGI_XPC_NOTIFY IRQ.
1043 */
1044
1045/*
1046 * Send an IPI to the remote partition that is associated with the
1047 * specified channel.
1048 */
1049#define XPC_NOTIFY_IRQ_SEND(_ch, _ipi_f, _irq_f) \
1050 xpc_notify_IRQ_send(_ch, _ipi_f, #_ipi_f, _irq_f)
1051
1052static inline void
1053xpc_notify_IRQ_send(struct xpc_channel *ch, u8 ipi_flag, char *ipi_flag_string,
1054 unsigned long *irq_flags)
1055{
1056 struct xpc_partition *part = &xpc_partitions[ch->partid];
1057 enum xpc_retval ret;
1058
1059
1060 if (likely(part->act_state != XPC_P_DEACTIVATING)) {
1061 ret = xpc_IPI_send(part->remote_IPI_amo_va,
1062 (u64) ipi_flag << (ch->number * 8),
1063 part->remote_IPI_nasid,
1064 part->remote_IPI_phys_cpuid,
1065 SGI_XPC_NOTIFY);
1066 dev_dbg(xpc_chan, "%s sent to partid=%d, channel=%d, ret=%d\n",
1067 ipi_flag_string, ch->partid, ch->number, ret);
1068 if (unlikely(ret != xpcSuccess)) {
1069 if (irq_flags != NULL) {
1070 spin_unlock_irqrestore(&ch->lock, *irq_flags);
1071 }
1072 XPC_DEACTIVATE_PARTITION(part, ret);
1073 if (irq_flags != NULL) {
1074 spin_lock_irqsave(&ch->lock, *irq_flags);
1075 }
1076 }
1077 }
1078}
1079
1080
1081/*
1082 * Make it look like the remote partition, which is associated with the
1083 * specified channel, sent us an IPI. This faked IPI will be handled
1084 * by xpc_dropped_IPI_check().
1085 */
1086#define XPC_NOTIFY_IRQ_SEND_LOCAL(_ch, _ipi_f) \
1087 xpc_notify_IRQ_send_local(_ch, _ipi_f, #_ipi_f)
1088
1089static inline void
1090xpc_notify_IRQ_send_local(struct xpc_channel *ch, u8 ipi_flag,
1091 char *ipi_flag_string)
1092{
1093 struct xpc_partition *part = &xpc_partitions[ch->partid];
1094
1095
1096 FETCHOP_STORE_OP(TO_AMO((u64) &part->local_IPI_amo_va->variable),
1097 FETCHOP_OR, ((u64) ipi_flag << (ch->number * 8)));
1098 dev_dbg(xpc_chan, "%s sent local from partid=%d, channel=%d\n",
1099 ipi_flag_string, ch->partid, ch->number);
1100}
1101
1102
1103/*
1104 * The sending and receiving of IPIs includes the setting of an AMO variable
1105 * to indicate the reason the IPI was sent. The 64-bit variable is divided
1106 * up into eight bytes, ordered from right to left. Byte zero pertains to
1107 * channel 0, byte one to channel 1, and so on. Each byte is described by
1108 * the following IPI flags.
1109 */
1110
1111#define XPC_IPI_CLOSEREQUEST 0x01
1112#define XPC_IPI_CLOSEREPLY 0x02
1113#define XPC_IPI_OPENREQUEST 0x04
1114#define XPC_IPI_OPENREPLY 0x08
1115#define XPC_IPI_MSGREQUEST 0x10
1116
1117
1118/* given an AMO variable and a channel#, get its associated IPI flags */
1119#define XPC_GET_IPI_FLAGS(_amo, _c) ((u8) (((_amo) >> ((_c) * 8)) & 0xff))
1120#define XPC_SET_IPI_FLAGS(_amo, _c, _f) (_amo) |= ((u64) (_f) << ((_c) * 8))
1121
1122#define XPC_ANY_OPENCLOSE_IPI_FLAGS_SET(_amo) ((_amo) & 0x0f0f0f0f0f0f0f0f)
1123#define XPC_ANY_MSG_IPI_FLAGS_SET(_amo) ((_amo) & 0x1010101010101010)
1124
1125
1126static inline void
1127xpc_IPI_send_closerequest(struct xpc_channel *ch, unsigned long *irq_flags)
1128{
1129 struct xpc_openclose_args *args = ch->local_openclose_args;
1130
1131
1132 args->reason = ch->reason;
1133
1134 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREQUEST, irq_flags);
1135}
1136
1137static inline void
1138xpc_IPI_send_closereply(struct xpc_channel *ch, unsigned long *irq_flags)
1139{
1140 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_CLOSEREPLY, irq_flags);
1141}
1142
1143static inline void
1144xpc_IPI_send_openrequest(struct xpc_channel *ch, unsigned long *irq_flags)
1145{
1146 struct xpc_openclose_args *args = ch->local_openclose_args;
1147
1148
1149 args->msg_size = ch->msg_size;
1150 args->local_nentries = ch->local_nentries;
1151
1152 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREQUEST, irq_flags);
1153}
1154
1155static inline void
1156xpc_IPI_send_openreply(struct xpc_channel *ch, unsigned long *irq_flags)
1157{
1158 struct xpc_openclose_args *args = ch->local_openclose_args;
1159
1160
1161 args->remote_nentries = ch->remote_nentries;
1162 args->local_nentries = ch->local_nentries;
1163 args->local_msgqueue_pa = __pa(ch->local_msgqueue);
1164
1165 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_OPENREPLY, irq_flags);
1166}
1167
1168static inline void
1169xpc_IPI_send_msgrequest(struct xpc_channel *ch)
1170{
1171 XPC_NOTIFY_IRQ_SEND(ch, XPC_IPI_MSGREQUEST, NULL);
1172}
1173
1174static inline void
1175xpc_IPI_send_local_msgrequest(struct xpc_channel *ch)
1176{
1177 XPC_NOTIFY_IRQ_SEND_LOCAL(ch, XPC_IPI_MSGREQUEST);
1178}
1179
1180
1181/*
1182 * Memory for XPC's AMO variables is allocated by the MSPEC driver. These
1183 * pages are located in the lowest granule. The lowest granule uses 4k pages
1184 * for cached references and an alternate TLB handler to never provide a
1185 * cacheable mapping for the entire region. This will prevent speculative
1186 * reading of cached copies of our lines from being issued which will cause
1187 * a PI FSB Protocol error to be generated by the SHUB. For XPC, we need 64
1188 * AMO variables (based on XP_MAX_PARTITIONS) for message notification and an
1189 * additional 128 AMO variables (based on XP_NASID_MASK_WORDS) for partition
1190 * activation and 2 AMO variables for partition deactivation.
1191 */
1192static inline AMO_t *
1193xpc_IPI_init(int index)
1194{
1195 AMO_t *amo = xpc_vars->amos_page + index;
1196
1197
1198 (void) xpc_IPI_receive(amo); /* clear AMO variable */
1199 return amo;
1200}
1201
1202
1203
1204static inline enum xpc_retval
1205xpc_map_bte_errors(bte_result_t error)
1206{
1207 switch (error) {
1208 case BTE_SUCCESS: return xpcSuccess;
1209 case BTEFAIL_DIR: return xpcBteDirectoryError;
1210 case BTEFAIL_POISON: return xpcBtePoisonError;
1211 case BTEFAIL_WERR: return xpcBteWriteError;
1212 case BTEFAIL_ACCESS: return xpcBteAccessError;
1213 case BTEFAIL_PWERR: return xpcBtePWriteError;
1214 case BTEFAIL_PRERR: return xpcBtePReadError;
1215 case BTEFAIL_TOUT: return xpcBteTimeOutError;
1216 case BTEFAIL_XTERR: return xpcBteXtalkError;
1217 case BTEFAIL_NOTAVAIL: return xpcBteNotAvailable;
1218 default: return xpcBteUnmappedError;
1219 }
1220}
1221
1222
1223
1224static inline void *
1225xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
1226{
1227 /* see if kmalloc will give us cachline aligned memory by default */
1228 *base = kmalloc(size, flags);
1229 if (*base == NULL) {
1230 return NULL;
1231 }
1232 if ((u64) *base == L1_CACHE_ALIGN((u64) *base)) {
1233 return *base;
1234 }
1235 kfree(*base);
1236
1237 /* nope, we'll have to do it ourselves */
1238 *base = kmalloc(size + L1_CACHE_BYTES, flags);
1239 if (*base == NULL) {
1240 return NULL;
1241 }
1242 return (void *) L1_CACHE_ALIGN((u64) *base);
1243}
1244
1245
1246/*
1247 * Check to see if there is any channel activity to/from the specified
1248 * partition.
1249 */
1250static inline void
1251xpc_check_for_channel_activity(struct xpc_partition *part)
1252{
1253 u64 IPI_amo;
1254 unsigned long irq_flags;
1255
1256
1257 IPI_amo = xpc_IPI_receive(part->local_IPI_amo_va);
1258 if (IPI_amo == 0) {
1259 return;
1260 }
1261
1262 spin_lock_irqsave(&part->IPI_lock, irq_flags);
1263 part->local_IPI_amo |= IPI_amo;
1264 spin_unlock_irqrestore(&part->IPI_lock, irq_flags);
1265
1266 dev_dbg(xpc_chan, "received IPI from partid=%d, IPI_amo=0x%lx\n",
1267 XPC_PARTID(part), IPI_amo);
1268
1269 xpc_wakeup_channel_mgr(part);
1270}
1271
1272
1273#endif /* _ASM_IA64_SN_XPC_H */
1274
diff --git a/include/asm-ia64/thread_info.h b/include/asm-ia64/thread_info.h
index 653bb7f9a753..1d6518fe1f02 100644
--- a/include/asm-ia64/thread_info.h
+++ b/include/asm-ia64/thread_info.h
@@ -93,6 +93,7 @@ struct thread_info {
93#define TIF_POLLING_NRFLAG 16 /* true if poll_idle() is polling TIF_NEED_RESCHED */ 93#define TIF_POLLING_NRFLAG 16 /* true if poll_idle() is polling TIF_NEED_RESCHED */
94#define TIF_MEMDIE 17 94#define TIF_MEMDIE 17
95#define TIF_MCA_INIT 18 /* this task is processing MCA or INIT */ 95#define TIF_MCA_INIT 18 /* this task is processing MCA or INIT */
96#define TIF_DB_DISABLED 19 /* debug trap disabled for fsyscall */
96 97
97#define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE) 98#define _TIF_SYSCALL_TRACE (1 << TIF_SYSCALL_TRACE)
98#define _TIF_SYSCALL_AUDIT (1 << TIF_SYSCALL_AUDIT) 99#define _TIF_SYSCALL_AUDIT (1 << TIF_SYSCALL_AUDIT)
@@ -100,9 +101,10 @@ struct thread_info {
100#define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME) 101#define _TIF_NOTIFY_RESUME (1 << TIF_NOTIFY_RESUME)
101#define _TIF_SIGPENDING (1 << TIF_SIGPENDING) 102#define _TIF_SIGPENDING (1 << TIF_SIGPENDING)
102#define _TIF_NEED_RESCHED (1 << TIF_NEED_RESCHED) 103#define _TIF_NEED_RESCHED (1 << TIF_NEED_RESCHED)
103#define _TIF_SIGDELAYED (1 << TIF_SIGDELAYED) 104#define _TIF_SIGDELAYED (1 << TIF_SIGDELAYED)
104#define _TIF_POLLING_NRFLAG (1 << TIF_POLLING_NRFLAG) 105#define _TIF_POLLING_NRFLAG (1 << TIF_POLLING_NRFLAG)
105#define _TIF_MCA_INIT (1 << TIF_MCA_INIT) 106#define _TIF_MCA_INIT (1 << TIF_MCA_INIT)
107#define _TIF_DB_DISABLED (1 << TIF_DB_DISABLED)
106 108
107/* "work to do on user-return" bits */ 109/* "work to do on user-return" bits */
108#define TIF_ALLWORK_MASK (_TIF_NOTIFY_RESUME|_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SIGDELAYED) 110#define TIF_ALLWORK_MASK (_TIF_NOTIFY_RESUME|_TIF_SIGPENDING|_TIF_NEED_RESCHED|_TIF_SYSCALL_TRACE|_TIF_SYSCALL_AUDIT|_TIF_SIGDELAYED)