Merge branch 'ima-memory-use-fixes'
[cascardo/linux.git] / drivers / misc / sgi-xp / xpc_uv.c
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) 2008-2009 Silicon Graphics, Inc.  All Rights Reserved.
7  */
8
9 /*
10  * Cross Partition Communication (XPC) uv-based functions.
11  *
12  *     Architecture specific implementation of common functions.
13  *
14  */
15
16 #include <linux/kernel.h>
17 #include <linux/mm.h>
18 #include <linux/interrupt.h>
19 #include <linux/delay.h>
20 #include <linux/device.h>
21 #include <linux/err.h>
22 #include <linux/slab.h>
23 #include <asm/uv/uv_hub.h>
24 #if defined CONFIG_X86_64
25 #include <asm/uv/bios.h>
26 #include <asm/uv/uv_irq.h>
27 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
28 #include <asm/sn/intr.h>
29 #include <asm/sn/sn_sal.h>
30 #endif
31 #include "../sgi-gru/gru.h"
32 #include "../sgi-gru/grukservices.h"
33 #include "xpc.h"
34
35 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
36 struct uv_IO_APIC_route_entry {
37         __u64   vector          :  8,
38                 delivery_mode   :  3,
39                 dest_mode       :  1,
40                 delivery_status :  1,
41                 polarity        :  1,
42                 __reserved_1    :  1,
43                 trigger         :  1,
44                 mask            :  1,
45                 __reserved_2    : 15,
46                 dest            : 32;
47 };
48 #endif
49
50 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
51
52 #define XPC_ACTIVATE_MSG_SIZE_UV        (1 * GRU_CACHE_LINE_BYTES)
53 #define XPC_ACTIVATE_MQ_SIZE_UV         (4 * XP_MAX_NPARTITIONS_UV * \
54                                          XPC_ACTIVATE_MSG_SIZE_UV)
55 #define XPC_ACTIVATE_IRQ_NAME           "xpc_activate"
56
57 #define XPC_NOTIFY_MSG_SIZE_UV          (2 * GRU_CACHE_LINE_BYTES)
58 #define XPC_NOTIFY_MQ_SIZE_UV           (4 * XP_MAX_NPARTITIONS_UV * \
59                                          XPC_NOTIFY_MSG_SIZE_UV)
60 #define XPC_NOTIFY_IRQ_NAME             "xpc_notify"
61
62 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
63 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
64
65 static int
66 xpc_setup_partitions_uv(void)
67 {
68         short partid;
69         struct xpc_partition_uv *part_uv;
70
71         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
72                 part_uv = &xpc_partitions[partid].sn.uv;
73
74                 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
75                 spin_lock_init(&part_uv->flags_lock);
76                 part_uv->remote_act_state = XPC_P_AS_INACTIVE;
77         }
78         return 0;
79 }
80
81 static void
82 xpc_teardown_partitions_uv(void)
83 {
84         short partid;
85         struct xpc_partition_uv *part_uv;
86         unsigned long irq_flags;
87
88         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
89                 part_uv = &xpc_partitions[partid].sn.uv;
90
91                 if (part_uv->cached_activate_gru_mq_desc != NULL) {
92                         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
93                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
94                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
95                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
96                         kfree(part_uv->cached_activate_gru_mq_desc);
97                         part_uv->cached_activate_gru_mq_desc = NULL;
98                         mutex_unlock(&part_uv->
99                                      cached_activate_gru_mq_desc_mutex);
100                 }
101         }
102 }
103
104 static int
105 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
106 {
107         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
108
109 #if defined CONFIG_X86_64
110         mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
111                         UV_AFFINITY_CPU);
112         if (mq->irq < 0) {
113                 dev_err(xpc_part, "uv_setup_irq() returned error=%d\n",
114                         -mq->irq);
115                 return mq->irq;
116         }
117
118         mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
119
120 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
121         if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
122                 mq->irq = SGI_XPC_ACTIVATE;
123         else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
124                 mq->irq = SGI_XPC_NOTIFY;
125         else
126                 return -EINVAL;
127
128         mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
129         uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
130 #else
131         #error not a supported configuration
132 #endif
133
134         return 0;
135 }
136
137 static void
138 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
139 {
140 #if defined CONFIG_X86_64
141         uv_teardown_irq(mq->irq);
142
143 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
144         int mmr_pnode;
145         unsigned long mmr_value;
146
147         mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
148         mmr_value = 1UL << 16;
149
150         uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
151 #else
152         #error not a supported configuration
153 #endif
154 }
155
156 static int
157 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
158 {
159         int ret;
160
161 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
162         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
163
164         ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address),
165                                     mq->order, &mq->mmr_offset);
166         if (ret < 0) {
167                 dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
168                         ret);
169                 return -EBUSY;
170         }
171 #elif defined CONFIG_X86_64
172         ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
173                                          mq->order, &mq->mmr_offset);
174         if (ret < 0) {
175                 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
176                         "ret=%d\n", ret);
177                 return ret;
178         }
179 #else
180         #error not a supported configuration
181 #endif
182
183         mq->watchlist_num = ret;
184         return 0;
185 }
186
187 static void
188 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
189 {
190         int ret;
191         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
192
193 #if defined CONFIG_X86_64
194         ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
195         BUG_ON(ret != BIOS_STATUS_SUCCESS);
196 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
197         ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
198         BUG_ON(ret != SALRET_OK);
199 #else
200         #error not a supported configuration
201 #endif
202 }
203
204 static struct xpc_gru_mq_uv *
205 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
206                      irq_handler_t irq_handler)
207 {
208         enum xp_retval xp_ret;
209         int ret;
210         int nid;
211         int nasid;
212         int pg_order;
213         struct page *page;
214         struct xpc_gru_mq_uv *mq;
215         struct uv_IO_APIC_route_entry *mmr_value;
216
217         mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
218         if (mq == NULL) {
219                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
220                         "a xpc_gru_mq_uv structure\n");
221                 ret = -ENOMEM;
222                 goto out_0;
223         }
224
225         mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
226                                   GFP_KERNEL);
227         if (mq->gru_mq_desc == NULL) {
228                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
229                         "a gru_message_queue_desc structure\n");
230                 ret = -ENOMEM;
231                 goto out_1;
232         }
233
234         pg_order = get_order(mq_size);
235         mq->order = pg_order + PAGE_SHIFT;
236         mq_size = 1UL << mq->order;
237
238         mq->mmr_blade = uv_cpu_to_blade_id(cpu);
239
240         nid = cpu_to_node(cpu);
241         page = alloc_pages_exact_node(nid, GFP_KERNEL | __GFP_ZERO | GFP_THISNODE,
242                                 pg_order);
243         if (page == NULL) {
244                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
245                         "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
246                 ret = -ENOMEM;
247                 goto out_2;
248         }
249         mq->address = page_address(page);
250
251         /* enable generation of irq when GRU mq operation occurs to this mq */
252         ret = xpc_gru_mq_watchlist_alloc_uv(mq);
253         if (ret != 0)
254                 goto out_3;
255
256         ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
257         if (ret != 0)
258                 goto out_4;
259
260         ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
261         if (ret != 0) {
262                 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
263                         mq->irq, -ret);
264                 goto out_5;
265         }
266
267         nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
268
269         mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
270         ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
271                                      nasid, mmr_value->vector, mmr_value->dest);
272         if (ret != 0) {
273                 dev_err(xpc_part, "gru_create_message_queue() returned "
274                         "error=%d\n", ret);
275                 ret = -EINVAL;
276                 goto out_6;
277         }
278
279         /* allow other partitions to access this GRU mq */
280         xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
281         if (xp_ret != xpSuccess) {
282                 ret = -EACCES;
283                 goto out_6;
284         }
285
286         return mq;
287
288         /* something went wrong */
289 out_6:
290         free_irq(mq->irq, NULL);
291 out_5:
292         xpc_release_gru_mq_irq_uv(mq);
293 out_4:
294         xpc_gru_mq_watchlist_free_uv(mq);
295 out_3:
296         free_pages((unsigned long)mq->address, pg_order);
297 out_2:
298         kfree(mq->gru_mq_desc);
299 out_1:
300         kfree(mq);
301 out_0:
302         return ERR_PTR(ret);
303 }
304
305 static void
306 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
307 {
308         unsigned int mq_size;
309         int pg_order;
310         int ret;
311
312         /* disallow other partitions to access GRU mq */
313         mq_size = 1UL << mq->order;
314         ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
315         BUG_ON(ret != xpSuccess);
316
317         /* unregister irq handler and release mq irq/vector mapping */
318         free_irq(mq->irq, NULL);
319         xpc_release_gru_mq_irq_uv(mq);
320
321         /* disable generation of irq when GRU mq op occurs to this mq */
322         xpc_gru_mq_watchlist_free_uv(mq);
323
324         pg_order = mq->order - PAGE_SHIFT;
325         free_pages((unsigned long)mq->address, pg_order);
326
327         kfree(mq);
328 }
329
330 static enum xp_retval
331 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
332                  size_t msg_size)
333 {
334         enum xp_retval xp_ret;
335         int ret;
336
337         while (1) {
338                 ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
339                 if (ret == MQE_OK) {
340                         xp_ret = xpSuccess;
341                         break;
342                 }
343
344                 if (ret == MQE_QUEUE_FULL) {
345                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
346                                 "error=MQE_QUEUE_FULL\n");
347                         /* !!! handle QLimit reached; delay & try again */
348                         /* ??? Do we add a limit to the number of retries? */
349                         (void)msleep_interruptible(10);
350                 } else if (ret == MQE_CONGESTION) {
351                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
352                                 "error=MQE_CONGESTION\n");
353                         /* !!! handle LB Overflow; simply try again */
354                         /* ??? Do we add a limit to the number of retries? */
355                 } else {
356                         /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
357                         dev_err(xpc_chan, "gru_send_message_gpa() returned "
358                                 "error=%d\n", ret);
359                         xp_ret = xpGruSendMqError;
360                         break;
361                 }
362         }
363         return xp_ret;
364 }
365
366 static void
367 xpc_process_activate_IRQ_rcvd_uv(void)
368 {
369         unsigned long irq_flags;
370         short partid;
371         struct xpc_partition *part;
372         u8 act_state_req;
373
374         DBUG_ON(xpc_activate_IRQ_rcvd == 0);
375
376         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
377         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
378                 part = &xpc_partitions[partid];
379
380                 if (part->sn.uv.act_state_req == 0)
381                         continue;
382
383                 xpc_activate_IRQ_rcvd--;
384                 BUG_ON(xpc_activate_IRQ_rcvd < 0);
385
386                 act_state_req = part->sn.uv.act_state_req;
387                 part->sn.uv.act_state_req = 0;
388                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
389
390                 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
391                         if (part->act_state == XPC_P_AS_INACTIVE)
392                                 xpc_activate_partition(part);
393                         else if (part->act_state == XPC_P_AS_DEACTIVATING)
394                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
395
396                 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
397                         if (part->act_state == XPC_P_AS_INACTIVE)
398                                 xpc_activate_partition(part);
399                         else
400                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
401
402                 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
403                         XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
404
405                 } else {
406                         BUG();
407                 }
408
409                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
410                 if (xpc_activate_IRQ_rcvd == 0)
411                         break;
412         }
413         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
414
415 }
416
417 static void
418 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
419                               struct xpc_activate_mq_msghdr_uv *msg_hdr,
420                               int *wakeup_hb_checker)
421 {
422         unsigned long irq_flags;
423         struct xpc_partition_uv *part_uv = &part->sn.uv;
424         struct xpc_openclose_args *args;
425
426         part_uv->remote_act_state = msg_hdr->act_state;
427
428         switch (msg_hdr->type) {
429         case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
430                 /* syncing of remote_act_state was just done above */
431                 break;
432
433         case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
434                 struct xpc_activate_mq_msg_activate_req_uv *msg;
435
436                 /*
437                  * ??? Do we deal here with ts_jiffies being different
438                  * ??? if act_state != XPC_P_AS_INACTIVE instead of
439                  * ??? below?
440                  */
441                 msg = container_of(msg_hdr, struct
442                                    xpc_activate_mq_msg_activate_req_uv, hdr);
443
444                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
445                 if (part_uv->act_state_req == 0)
446                         xpc_activate_IRQ_rcvd++;
447                 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
448                 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
449                 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
450                 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
451
452                 if (msg->activate_gru_mq_desc_gpa !=
453                     part_uv->activate_gru_mq_desc_gpa) {
454                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
455                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
456                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
457                         part_uv->activate_gru_mq_desc_gpa =
458                             msg->activate_gru_mq_desc_gpa;
459                 }
460                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
461
462                 (*wakeup_hb_checker)++;
463                 break;
464         }
465         case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
466                 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
467
468                 msg = container_of(msg_hdr, struct
469                                    xpc_activate_mq_msg_deactivate_req_uv, hdr);
470
471                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
472                 if (part_uv->act_state_req == 0)
473                         xpc_activate_IRQ_rcvd++;
474                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
475                 part_uv->reason = msg->reason;
476                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
477
478                 (*wakeup_hb_checker)++;
479                 return;
480         }
481         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
482                 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
483
484                 msg = container_of(msg_hdr, struct
485                                    xpc_activate_mq_msg_chctl_closerequest_uv,
486                                    hdr);
487                 args = &part->remote_openclose_args[msg->ch_number];
488                 args->reason = msg->reason;
489
490                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
491                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
492                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
493
494                 xpc_wakeup_channel_mgr(part);
495                 break;
496         }
497         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
498                 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
499
500                 msg = container_of(msg_hdr, struct
501                                    xpc_activate_mq_msg_chctl_closereply_uv,
502                                    hdr);
503
504                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
505                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
506                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
507
508                 xpc_wakeup_channel_mgr(part);
509                 break;
510         }
511         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
512                 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
513
514                 msg = container_of(msg_hdr, struct
515                                    xpc_activate_mq_msg_chctl_openrequest_uv,
516                                    hdr);
517                 args = &part->remote_openclose_args[msg->ch_number];
518                 args->entry_size = msg->entry_size;
519                 args->local_nentries = msg->local_nentries;
520
521                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
522                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
523                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
524
525                 xpc_wakeup_channel_mgr(part);
526                 break;
527         }
528         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
529                 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
530
531                 msg = container_of(msg_hdr, struct
532                                    xpc_activate_mq_msg_chctl_openreply_uv, hdr);
533                 args = &part->remote_openclose_args[msg->ch_number];
534                 args->remote_nentries = msg->remote_nentries;
535                 args->local_nentries = msg->local_nentries;
536                 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
537
538                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
539                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
540                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
541
542                 xpc_wakeup_channel_mgr(part);
543                 break;
544         }
545         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
546                 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
547
548                 msg = container_of(msg_hdr, struct
549                                 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
550                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
551                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
552                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
553
554                 xpc_wakeup_channel_mgr(part);
555         }
556         case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
557                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
558                 part_uv->flags |= XPC_P_ENGAGED_UV;
559                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
560                 break;
561
562         case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
563                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
564                 part_uv->flags &= ~XPC_P_ENGAGED_UV;
565                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
566                 break;
567
568         default:
569                 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
570                         "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
571
572                 /* get hb checker to deactivate from the remote partition */
573                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
574                 if (part_uv->act_state_req == 0)
575                         xpc_activate_IRQ_rcvd++;
576                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
577                 part_uv->reason = xpBadMsgType;
578                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
579
580                 (*wakeup_hb_checker)++;
581                 return;
582         }
583
584         if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
585             part->remote_rp_ts_jiffies != 0) {
586                 /*
587                  * ??? Does what we do here need to be sensitive to
588                  * ??? act_state or remote_act_state?
589                  */
590                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
591                 if (part_uv->act_state_req == 0)
592                         xpc_activate_IRQ_rcvd++;
593                 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
594                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
595
596                 (*wakeup_hb_checker)++;
597         }
598 }
599
600 static irqreturn_t
601 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
602 {
603         struct xpc_activate_mq_msghdr_uv *msg_hdr;
604         short partid;
605         struct xpc_partition *part;
606         int wakeup_hb_checker = 0;
607         int part_referenced;
608
609         while (1) {
610                 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
611                 if (msg_hdr == NULL)
612                         break;
613
614                 partid = msg_hdr->partid;
615                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
616                         dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
617                                 "received invalid partid=0x%x in message\n",
618                                 partid);
619                 } else {
620                         part = &xpc_partitions[partid];
621
622                         part_referenced = xpc_part_ref(part);
623                         xpc_handle_activate_mq_msg_uv(part, msg_hdr,
624                                                       &wakeup_hb_checker);
625                         if (part_referenced)
626                                 xpc_part_deref(part);
627                 }
628
629                 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
630         }
631
632         if (wakeup_hb_checker)
633                 wake_up_interruptible(&xpc_activate_IRQ_wq);
634
635         return IRQ_HANDLED;
636 }
637
638 static enum xp_retval
639 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
640                                 unsigned long gru_mq_desc_gpa)
641 {
642         enum xp_retval ret;
643
644         ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
645                                sizeof(struct gru_message_queue_desc));
646         if (ret == xpSuccess)
647                 gru_mq_desc->mq = NULL;
648
649         return ret;
650 }
651
652 static enum xp_retval
653 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
654                          int msg_type)
655 {
656         struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
657         struct xpc_partition_uv *part_uv = &part->sn.uv;
658         struct gru_message_queue_desc *gru_mq_desc;
659         unsigned long irq_flags;
660         enum xp_retval ret;
661
662         DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
663
664         msg_hdr->type = msg_type;
665         msg_hdr->partid = xp_partition_id;
666         msg_hdr->act_state = part->act_state;
667         msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
668
669         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
670 again:
671         if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
672                 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
673                 if (gru_mq_desc == NULL) {
674                         gru_mq_desc = kmalloc(sizeof(struct
675                                               gru_message_queue_desc),
676                                               GFP_KERNEL);
677                         if (gru_mq_desc == NULL) {
678                                 ret = xpNoMemory;
679                                 goto done;
680                         }
681                         part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
682                 }
683
684                 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
685                                                       part_uv->
686                                                       activate_gru_mq_desc_gpa);
687                 if (ret != xpSuccess)
688                         goto done;
689
690                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
691                 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
692                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
693         }
694
695         /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
696         ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
697                                msg_size);
698         if (ret != xpSuccess) {
699                 smp_rmb();      /* ensure a fresh copy of part_uv->flags */
700                 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
701                         goto again;
702         }
703 done:
704         mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
705         return ret;
706 }
707
708 static void
709 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
710                               size_t msg_size, int msg_type)
711 {
712         enum xp_retval ret;
713
714         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
715         if (unlikely(ret != xpSuccess))
716                 XPC_DEACTIVATE_PARTITION(part, ret);
717 }
718
719 static void
720 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
721                          void *msg, size_t msg_size, int msg_type)
722 {
723         struct xpc_partition *part = &xpc_partitions[ch->partid];
724         enum xp_retval ret;
725
726         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
727         if (unlikely(ret != xpSuccess)) {
728                 if (irq_flags != NULL)
729                         spin_unlock_irqrestore(&ch->lock, *irq_flags);
730
731                 XPC_DEACTIVATE_PARTITION(part, ret);
732
733                 if (irq_flags != NULL)
734                         spin_lock_irqsave(&ch->lock, *irq_flags);
735         }
736 }
737
738 static void
739 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
740 {
741         unsigned long irq_flags;
742         struct xpc_partition_uv *part_uv = &part->sn.uv;
743
744         /*
745          * !!! Make our side think that the remote partition sent an activate
746          * !!! mq message our way by doing what the activate IRQ handler would
747          * !!! do had one really been sent.
748          */
749
750         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
751         if (part_uv->act_state_req == 0)
752                 xpc_activate_IRQ_rcvd++;
753         part_uv->act_state_req = act_state_req;
754         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
755
756         wake_up_interruptible(&xpc_activate_IRQ_wq);
757 }
758
759 static enum xp_retval
760 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
761                                   size_t *len)
762 {
763         s64 status;
764         enum xp_retval ret;
765
766 #if defined CONFIG_X86_64
767         status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
768                                           (u64 *)len);
769         if (status == BIOS_STATUS_SUCCESS)
770                 ret = xpSuccess;
771         else if (status == BIOS_STATUS_MORE_PASSES)
772                 ret = xpNeedMoreInfo;
773         else
774                 ret = xpBiosError;
775
776 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
777         status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
778         if (status == SALRET_OK)
779                 ret = xpSuccess;
780         else if (status == SALRET_MORE_PASSES)
781                 ret = xpNeedMoreInfo;
782         else
783                 ret = xpSalError;
784
785 #else
786         #error not a supported configuration
787 #endif
788
789         return ret;
790 }
791
792 static int
793 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
794 {
795         xpc_heartbeat_uv =
796             &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
797         rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
798         rp->sn.uv.activate_gru_mq_desc_gpa =
799             uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
800         return 0;
801 }
802
803 static void
804 xpc_allow_hb_uv(short partid)
805 {
806 }
807
808 static void
809 xpc_disallow_hb_uv(short partid)
810 {
811 }
812
813 static void
814 xpc_disallow_all_hbs_uv(void)
815 {
816 }
817
818 static void
819 xpc_increment_heartbeat_uv(void)
820 {
821         xpc_heartbeat_uv->value++;
822 }
823
824 static void
825 xpc_offline_heartbeat_uv(void)
826 {
827         xpc_increment_heartbeat_uv();
828         xpc_heartbeat_uv->offline = 1;
829 }
830
831 static void
832 xpc_online_heartbeat_uv(void)
833 {
834         xpc_increment_heartbeat_uv();
835         xpc_heartbeat_uv->offline = 0;
836 }
837
838 static void
839 xpc_heartbeat_init_uv(void)
840 {
841         xpc_heartbeat_uv->value = 1;
842         xpc_heartbeat_uv->offline = 0;
843 }
844
845 static void
846 xpc_heartbeat_exit_uv(void)
847 {
848         xpc_offline_heartbeat_uv();
849 }
850
851 static enum xp_retval
852 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
853 {
854         struct xpc_partition_uv *part_uv = &part->sn.uv;
855         enum xp_retval ret;
856
857         ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
858                                part_uv->heartbeat_gpa,
859                                sizeof(struct xpc_heartbeat_uv));
860         if (ret != xpSuccess)
861                 return ret;
862
863         if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
864             !part_uv->cached_heartbeat.offline) {
865
866                 ret = xpNoHeartbeat;
867         } else {
868                 part->last_heartbeat = part_uv->cached_heartbeat.value;
869         }
870         return ret;
871 }
872
873 static void
874 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
875                                     unsigned long remote_rp_gpa, int nasid)
876 {
877         short partid = remote_rp->SAL_partid;
878         struct xpc_partition *part = &xpc_partitions[partid];
879         struct xpc_activate_mq_msg_activate_req_uv msg;
880
881         part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
882         part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
883         part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
884         part->sn.uv.activate_gru_mq_desc_gpa =
885             remote_rp->sn.uv.activate_gru_mq_desc_gpa;
886
887         /*
888          * ??? Is it a good idea to make this conditional on what is
889          * ??? potentially stale state information?
890          */
891         if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
892                 msg.rp_gpa = uv_gpa(xpc_rsvd_page);
893                 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
894                 msg.activate_gru_mq_desc_gpa =
895                     xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
896                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
897                                            XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
898         }
899
900         if (part->act_state == XPC_P_AS_INACTIVE)
901                 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
902 }
903
904 static void
905 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
906 {
907         xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
908 }
909
910 static void
911 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
912 {
913         struct xpc_activate_mq_msg_deactivate_req_uv msg;
914
915         /*
916          * ??? Is it a good idea to make this conditional on what is
917          * ??? potentially stale state information?
918          */
919         if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
920             part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
921
922                 msg.reason = part->reason;
923                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
924                                          XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
925         }
926 }
927
928 static void
929 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
930 {
931         /* nothing needs to be done */
932         return;
933 }
934
935 static void
936 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
937 {
938         head->first = NULL;
939         head->last = NULL;
940         spin_lock_init(&head->lock);
941         head->n_entries = 0;
942 }
943
944 static void *
945 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
946 {
947         unsigned long irq_flags;
948         struct xpc_fifo_entry_uv *first;
949
950         spin_lock_irqsave(&head->lock, irq_flags);
951         first = head->first;
952         if (head->first != NULL) {
953                 head->first = first->next;
954                 if (head->first == NULL)
955                         head->last = NULL;
956
957                 head->n_entries--;
958                 BUG_ON(head->n_entries < 0);
959
960                 first->next = NULL;
961         }
962         spin_unlock_irqrestore(&head->lock, irq_flags);
963         return first;
964 }
965
966 static void
967 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
968                       struct xpc_fifo_entry_uv *last)
969 {
970         unsigned long irq_flags;
971
972         last->next = NULL;
973         spin_lock_irqsave(&head->lock, irq_flags);
974         if (head->last != NULL)
975                 head->last->next = last;
976         else
977                 head->first = last;
978         head->last = last;
979         head->n_entries++;
980         spin_unlock_irqrestore(&head->lock, irq_flags);
981 }
982
983 static int
984 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
985 {
986         return head->n_entries;
987 }
988
989 /*
990  * Setup the channel structures that are uv specific.
991  */
992 static enum xp_retval
993 xpc_setup_ch_structures_uv(struct xpc_partition *part)
994 {
995         struct xpc_channel_uv *ch_uv;
996         int ch_number;
997
998         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
999                 ch_uv = &part->channels[ch_number].sn.uv;
1000
1001                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1002                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1003         }
1004
1005         return xpSuccess;
1006 }
1007
1008 /*
1009  * Teardown the channel structures that are uv specific.
1010  */
1011 static void
1012 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1013 {
1014         /* nothing needs to be done */
1015         return;
1016 }
1017
1018 static enum xp_retval
1019 xpc_make_first_contact_uv(struct xpc_partition *part)
1020 {
1021         struct xpc_activate_mq_msg_uv msg;
1022
1023         /*
1024          * We send a sync msg to get the remote partition's remote_act_state
1025          * updated to our current act_state which at this point should
1026          * be XPC_P_AS_ACTIVATING.
1027          */
1028         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1029                                       XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1030
1031         while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
1032                  (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
1033
1034                 dev_dbg(xpc_part, "waiting to make first contact with "
1035                         "partition %d\n", XPC_PARTID(part));
1036
1037                 /* wait a 1/4 of a second or so */
1038                 (void)msleep_interruptible(250);
1039
1040                 if (part->act_state == XPC_P_AS_DEACTIVATING)
1041                         return part->reason;
1042         }
1043
1044         return xpSuccess;
1045 }
1046
1047 static u64
1048 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1049 {
1050         unsigned long irq_flags;
1051         union xpc_channel_ctl_flags chctl;
1052
1053         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1054         chctl = part->chctl;
1055         if (chctl.all_flags != 0)
1056                 part->chctl.all_flags = 0;
1057
1058         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1059         return chctl.all_flags;
1060 }
1061
1062 static enum xp_retval
1063 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1064 {
1065         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1066         struct xpc_send_msg_slot_uv *msg_slot;
1067         unsigned long irq_flags;
1068         int nentries;
1069         int entry;
1070         size_t nbytes;
1071
1072         for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1073                 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1074                 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1075                 if (ch_uv->send_msg_slots == NULL)
1076                         continue;
1077
1078                 for (entry = 0; entry < nentries; entry++) {
1079                         msg_slot = &ch_uv->send_msg_slots[entry];
1080
1081                         msg_slot->msg_slot_number = entry;
1082                         xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1083                                               &msg_slot->next);
1084                 }
1085
1086                 spin_lock_irqsave(&ch->lock, irq_flags);
1087                 if (nentries < ch->local_nentries)
1088                         ch->local_nentries = nentries;
1089                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1090                 return xpSuccess;
1091         }
1092
1093         return xpNoMemory;
1094 }
1095
1096 static enum xp_retval
1097 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1098 {
1099         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1100         struct xpc_notify_mq_msg_uv *msg_slot;
1101         unsigned long irq_flags;
1102         int nentries;
1103         int entry;
1104         size_t nbytes;
1105
1106         for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1107                 nbytes = nentries * ch->entry_size;
1108                 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1109                 if (ch_uv->recv_msg_slots == NULL)
1110                         continue;
1111
1112                 for (entry = 0; entry < nentries; entry++) {
1113                         msg_slot = ch_uv->recv_msg_slots +
1114                             entry * ch->entry_size;
1115
1116                         msg_slot->hdr.msg_slot_number = entry;
1117                 }
1118
1119                 spin_lock_irqsave(&ch->lock, irq_flags);
1120                 if (nentries < ch->remote_nentries)
1121                         ch->remote_nentries = nentries;
1122                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1123                 return xpSuccess;
1124         }
1125
1126         return xpNoMemory;
1127 }
1128
1129 /*
1130  * Allocate msg_slots associated with the channel.
1131  */
1132 static enum xp_retval
1133 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1134 {
1135         static enum xp_retval ret;
1136         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1137
1138         DBUG_ON(ch->flags & XPC_C_SETUP);
1139
1140         ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1141                                                    gru_message_queue_desc),
1142                                                    GFP_KERNEL);
1143         if (ch_uv->cached_notify_gru_mq_desc == NULL)
1144                 return xpNoMemory;
1145
1146         ret = xpc_allocate_send_msg_slot_uv(ch);
1147         if (ret == xpSuccess) {
1148
1149                 ret = xpc_allocate_recv_msg_slot_uv(ch);
1150                 if (ret != xpSuccess) {
1151                         kfree(ch_uv->send_msg_slots);
1152                         xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1153                 }
1154         }
1155         return ret;
1156 }
1157
1158 /*
1159  * Free up msg_slots and clear other stuff that were setup for the specified
1160  * channel.
1161  */
1162 static void
1163 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1164 {
1165         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1166
1167         DBUG_ON(!spin_is_locked(&ch->lock));
1168
1169         kfree(ch_uv->cached_notify_gru_mq_desc);
1170         ch_uv->cached_notify_gru_mq_desc = NULL;
1171
1172         if (ch->flags & XPC_C_SETUP) {
1173                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1174                 kfree(ch_uv->send_msg_slots);
1175                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1176                 kfree(ch_uv->recv_msg_slots);
1177         }
1178 }
1179
1180 static void
1181 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1182 {
1183         struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1184
1185         msg.ch_number = ch->number;
1186         msg.reason = ch->reason;
1187         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1188                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1189 }
1190
1191 static void
1192 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1193 {
1194         struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1195
1196         msg.ch_number = ch->number;
1197         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1198                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1199 }
1200
1201 static void
1202 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1203 {
1204         struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1205
1206         msg.ch_number = ch->number;
1207         msg.entry_size = ch->entry_size;
1208         msg.local_nentries = ch->local_nentries;
1209         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1210                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1211 }
1212
1213 static void
1214 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1215 {
1216         struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1217
1218         msg.ch_number = ch->number;
1219         msg.local_nentries = ch->local_nentries;
1220         msg.remote_nentries = ch->remote_nentries;
1221         msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1222         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1223                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1224 }
1225
1226 static void
1227 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1228 {
1229         struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1230
1231         msg.ch_number = ch->number;
1232         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1233                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1234 }
1235
1236 static void
1237 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1238 {
1239         unsigned long irq_flags;
1240
1241         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1242         part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1243         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1244
1245         xpc_wakeup_channel_mgr(part);
1246 }
1247
1248 static enum xp_retval
1249 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1250                                unsigned long gru_mq_desc_gpa)
1251 {
1252         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1253
1254         DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1255         return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1256                                                gru_mq_desc_gpa);
1257 }
1258
1259 static void
1260 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1261 {
1262         struct xpc_activate_mq_msg_uv msg;
1263
1264         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1265                                       XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1266 }
1267
1268 static void
1269 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1270 {
1271         struct xpc_activate_mq_msg_uv msg;
1272
1273         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1274                                       XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1275 }
1276
1277 static void
1278 xpc_assume_partition_disengaged_uv(short partid)
1279 {
1280         struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1281         unsigned long irq_flags;
1282
1283         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1284         part_uv->flags &= ~XPC_P_ENGAGED_UV;
1285         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1286 }
1287
1288 static int
1289 xpc_partition_engaged_uv(short partid)
1290 {
1291         return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1292 }
1293
1294 static int
1295 xpc_any_partition_engaged_uv(void)
1296 {
1297         struct xpc_partition_uv *part_uv;
1298         short partid;
1299
1300         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1301                 part_uv = &xpc_partitions[partid].sn.uv;
1302                 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1303                         return 1;
1304         }
1305         return 0;
1306 }
1307
1308 static enum xp_retval
1309 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1310                          struct xpc_send_msg_slot_uv **address_of_msg_slot)
1311 {
1312         enum xp_retval ret;
1313         struct xpc_send_msg_slot_uv *msg_slot;
1314         struct xpc_fifo_entry_uv *entry;
1315
1316         while (1) {
1317                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1318                 if (entry != NULL)
1319                         break;
1320
1321                 if (flags & XPC_NOWAIT)
1322                         return xpNoWait;
1323
1324                 ret = xpc_allocate_msg_wait(ch);
1325                 if (ret != xpInterrupted && ret != xpTimeout)
1326                         return ret;
1327         }
1328
1329         msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1330         *address_of_msg_slot = msg_slot;
1331         return xpSuccess;
1332 }
1333
1334 static void
1335 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1336                      struct xpc_send_msg_slot_uv *msg_slot)
1337 {
1338         xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1339
1340         /* wakeup anyone waiting for a free msg slot */
1341         if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1342                 wake_up(&ch->msg_allocate_wq);
1343 }
1344
1345 static void
1346 xpc_notify_sender_uv(struct xpc_channel *ch,
1347                      struct xpc_send_msg_slot_uv *msg_slot,
1348                      enum xp_retval reason)
1349 {
1350         xpc_notify_func func = msg_slot->func;
1351
1352         if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1353
1354                 atomic_dec(&ch->n_to_notify);
1355
1356                 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1357                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1358                         msg_slot->msg_slot_number, ch->partid, ch->number);
1359
1360                 func(reason, ch->partid, ch->number, msg_slot->key);
1361
1362                 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1363                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1364                         msg_slot->msg_slot_number, ch->partid, ch->number);
1365         }
1366 }
1367
1368 static void
1369 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1370                             struct xpc_notify_mq_msg_uv *msg)
1371 {
1372         struct xpc_send_msg_slot_uv *msg_slot;
1373         int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1374
1375         msg_slot = &ch->sn.uv.send_msg_slots[entry];
1376
1377         BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1378         msg_slot->msg_slot_number += ch->local_nentries;
1379
1380         if (msg_slot->func != NULL)
1381                 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1382
1383         xpc_free_msg_slot_uv(ch, msg_slot);
1384 }
1385
1386 static void
1387 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1388                             struct xpc_notify_mq_msg_uv *msg)
1389 {
1390         struct xpc_partition_uv *part_uv = &part->sn.uv;
1391         struct xpc_channel *ch;
1392         struct xpc_channel_uv *ch_uv;
1393         struct xpc_notify_mq_msg_uv *msg_slot;
1394         unsigned long irq_flags;
1395         int ch_number = msg->hdr.ch_number;
1396
1397         if (unlikely(ch_number >= part->nchannels)) {
1398                 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1399                         "channel number=0x%x in message from partid=%d\n",
1400                         ch_number, XPC_PARTID(part));
1401
1402                 /* get hb checker to deactivate from the remote partition */
1403                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1404                 if (part_uv->act_state_req == 0)
1405                         xpc_activate_IRQ_rcvd++;
1406                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1407                 part_uv->reason = xpBadChannelNumber;
1408                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1409
1410                 wake_up_interruptible(&xpc_activate_IRQ_wq);
1411                 return;
1412         }
1413
1414         ch = &part->channels[ch_number];
1415         xpc_msgqueue_ref(ch);
1416
1417         if (!(ch->flags & XPC_C_CONNECTED)) {
1418                 xpc_msgqueue_deref(ch);
1419                 return;
1420         }
1421
1422         /* see if we're really dealing with an ACK for a previously sent msg */
1423         if (msg->hdr.size == 0) {
1424                 xpc_handle_notify_mq_ack_uv(ch, msg);
1425                 xpc_msgqueue_deref(ch);
1426                 return;
1427         }
1428
1429         /* we're dealing with a normal message sent via the notify_mq */
1430         ch_uv = &ch->sn.uv;
1431
1432         msg_slot = ch_uv->recv_msg_slots +
1433             (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1434
1435         BUG_ON(msg_slot->hdr.size != 0);
1436
1437         memcpy(msg_slot, msg, msg->hdr.size);
1438
1439         xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1440
1441         if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1442                 /*
1443                  * If there is an existing idle kthread get it to deliver
1444                  * the payload, otherwise we'll have to get the channel mgr
1445                  * for this partition to create a kthread to do the delivery.
1446                  */
1447                 if (atomic_read(&ch->kthreads_idle) > 0)
1448                         wake_up_nr(&ch->idle_wq, 1);
1449                 else
1450                         xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1451         }
1452         xpc_msgqueue_deref(ch);
1453 }
1454
1455 static irqreturn_t
1456 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1457 {
1458         struct xpc_notify_mq_msg_uv *msg;
1459         short partid;
1460         struct xpc_partition *part;
1461
1462         while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1463                NULL) {
1464
1465                 partid = msg->hdr.partid;
1466                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1467                         dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1468                                 "invalid partid=0x%x in message\n", partid);
1469                 } else {
1470                         part = &xpc_partitions[partid];
1471
1472                         if (xpc_part_ref(part)) {
1473                                 xpc_handle_notify_mq_msg_uv(part, msg);
1474                                 xpc_part_deref(part);
1475                         }
1476                 }
1477
1478                 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1479         }
1480
1481         return IRQ_HANDLED;
1482 }
1483
1484 static int
1485 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1486 {
1487         return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1488 }
1489
1490 static void
1491 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1492 {
1493         struct xpc_channel *ch = &part->channels[ch_number];
1494         int ndeliverable_payloads;
1495
1496         xpc_msgqueue_ref(ch);
1497
1498         ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1499
1500         if (ndeliverable_payloads > 0 &&
1501             (ch->flags & XPC_C_CONNECTED) &&
1502             (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1503
1504                 xpc_activate_kthreads(ch, ndeliverable_payloads);
1505         }
1506
1507         xpc_msgqueue_deref(ch);
1508 }
1509
1510 static enum xp_retval
1511 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1512                     u16 payload_size, u8 notify_type, xpc_notify_func func,
1513                     void *key)
1514 {
1515         enum xp_retval ret = xpSuccess;
1516         struct xpc_send_msg_slot_uv *msg_slot = NULL;
1517         struct xpc_notify_mq_msg_uv *msg;
1518         u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1519         size_t msg_size;
1520
1521         DBUG_ON(notify_type != XPC_N_CALL);
1522
1523         msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1524         if (msg_size > ch->entry_size)
1525                 return xpPayloadTooBig;
1526
1527         xpc_msgqueue_ref(ch);
1528
1529         if (ch->flags & XPC_C_DISCONNECTING) {
1530                 ret = ch->reason;
1531                 goto out_1;
1532         }
1533         if (!(ch->flags & XPC_C_CONNECTED)) {
1534                 ret = xpNotConnected;
1535                 goto out_1;
1536         }
1537
1538         ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1539         if (ret != xpSuccess)
1540                 goto out_1;
1541
1542         if (func != NULL) {
1543                 atomic_inc(&ch->n_to_notify);
1544
1545                 msg_slot->key = key;
1546                 smp_wmb(); /* a non-NULL func must hit memory after the key */
1547                 msg_slot->func = func;
1548
1549                 if (ch->flags & XPC_C_DISCONNECTING) {
1550                         ret = ch->reason;
1551                         goto out_2;
1552                 }
1553         }
1554
1555         msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1556         msg->hdr.partid = xp_partition_id;
1557         msg->hdr.ch_number = ch->number;
1558         msg->hdr.size = msg_size;
1559         msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1560         memcpy(&msg->payload, payload, payload_size);
1561
1562         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1563                                msg_size);
1564         if (ret == xpSuccess)
1565                 goto out_1;
1566
1567         XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1568 out_2:
1569         if (func != NULL) {
1570                 /*
1571                  * Try to NULL the msg_slot's func field. If we fail, then
1572                  * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1573                  * case we need to pretend we succeeded to send the message
1574                  * since the user will get a callout for the disconnect error
1575                  * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1576                  * error returned here will confuse them. Additionally, since
1577                  * in this case the channel is being disconnected we don't need
1578                  * to put the the msg_slot back on the free list.
1579                  */
1580                 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1581                         ret = xpSuccess;
1582                         goto out_1;
1583                 }
1584
1585                 msg_slot->key = NULL;
1586                 atomic_dec(&ch->n_to_notify);
1587         }
1588         xpc_free_msg_slot_uv(ch, msg_slot);
1589 out_1:
1590         xpc_msgqueue_deref(ch);
1591         return ret;
1592 }
1593
1594 /*
1595  * Tell the callers of xpc_send_notify() that the status of their payloads
1596  * is unknown because the channel is now disconnecting.
1597  *
1598  * We don't worry about putting these msg_slots on the free list since the
1599  * msg_slots themselves are about to be kfree'd.
1600  */
1601 static void
1602 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1603 {
1604         struct xpc_send_msg_slot_uv *msg_slot;
1605         int entry;
1606
1607         DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1608
1609         for (entry = 0; entry < ch->local_nentries; entry++) {
1610
1611                 if (atomic_read(&ch->n_to_notify) == 0)
1612                         break;
1613
1614                 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1615                 if (msg_slot->func != NULL)
1616                         xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1617         }
1618 }
1619
1620 /*
1621  * Get the next deliverable message's payload.
1622  */
1623 static void *
1624 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1625 {
1626         struct xpc_fifo_entry_uv *entry;
1627         struct xpc_notify_mq_msg_uv *msg;
1628         void *payload = NULL;
1629
1630         if (!(ch->flags & XPC_C_DISCONNECTING)) {
1631                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1632                 if (entry != NULL) {
1633                         msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1634                                            hdr.u.next);
1635                         payload = &msg->payload;
1636                 }
1637         }
1638         return payload;
1639 }
1640
1641 static void
1642 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1643 {
1644         struct xpc_notify_mq_msg_uv *msg;
1645         enum xp_retval ret;
1646
1647         msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1648
1649         /* return an ACK to the sender of this message */
1650
1651         msg->hdr.partid = xp_partition_id;
1652         msg->hdr.size = 0;      /* size of zero indicates this is an ACK */
1653
1654         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1655                                sizeof(struct xpc_notify_mq_msghdr_uv));
1656         if (ret != xpSuccess)
1657                 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1658 }
1659
1660 static struct xpc_arch_operations xpc_arch_ops_uv = {
1661         .setup_partitions = xpc_setup_partitions_uv,
1662         .teardown_partitions = xpc_teardown_partitions_uv,
1663         .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1664         .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1665         .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1666
1667         .allow_hb = xpc_allow_hb_uv,
1668         .disallow_hb = xpc_disallow_hb_uv,
1669         .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1670         .increment_heartbeat = xpc_increment_heartbeat_uv,
1671         .offline_heartbeat = xpc_offline_heartbeat_uv,
1672         .online_heartbeat = xpc_online_heartbeat_uv,
1673         .heartbeat_init = xpc_heartbeat_init_uv,
1674         .heartbeat_exit = xpc_heartbeat_exit_uv,
1675         .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1676
1677         .request_partition_activation =
1678                 xpc_request_partition_activation_uv,
1679         .request_partition_reactivation =
1680                 xpc_request_partition_reactivation_uv,
1681         .request_partition_deactivation =
1682                 xpc_request_partition_deactivation_uv,
1683         .cancel_partition_deactivation_request =
1684                 xpc_cancel_partition_deactivation_request_uv,
1685
1686         .setup_ch_structures = xpc_setup_ch_structures_uv,
1687         .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1688
1689         .make_first_contact = xpc_make_first_contact_uv,
1690
1691         .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1692         .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1693         .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1694         .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1695         .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1696         .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1697         .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1698
1699         .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1700
1701         .setup_msg_structures = xpc_setup_msg_structures_uv,
1702         .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1703
1704         .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1705         .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1706         .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1707         .partition_engaged = xpc_partition_engaged_uv,
1708         .any_partition_engaged = xpc_any_partition_engaged_uv,
1709
1710         .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1711         .send_payload = xpc_send_payload_uv,
1712         .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1713         .received_payload = xpc_received_payload_uv,
1714         .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1715 };
1716
1717 int
1718 xpc_init_uv(void)
1719 {
1720         xpc_arch_ops = xpc_arch_ops_uv;
1721
1722         if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1723                 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1724                         XPC_MSG_HDR_MAX_SIZE);
1725                 return -E2BIG;
1726         }
1727
1728         xpc_activate_mq_uv = xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, 0,
1729                                                   XPC_ACTIVATE_IRQ_NAME,
1730                                                   xpc_handle_activate_IRQ_uv);
1731         if (IS_ERR(xpc_activate_mq_uv))
1732                 return PTR_ERR(xpc_activate_mq_uv);
1733
1734         xpc_notify_mq_uv = xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, 0,
1735                                                 XPC_NOTIFY_IRQ_NAME,
1736                                                 xpc_handle_notify_IRQ_uv);
1737         if (IS_ERR(xpc_notify_mq_uv)) {
1738                 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1739                 return PTR_ERR(xpc_notify_mq_uv);
1740         }
1741
1742         return 0;
1743 }
1744
1745 void
1746 xpc_exit_uv(void)
1747 {
1748         xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1749         xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1750 }