Merge branch 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[pandora-kernel.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 part_setup,
421                               int *wakeup_hb_checker)
422 {
423         unsigned long irq_flags;
424         struct xpc_partition_uv *part_uv = &part->sn.uv;
425         struct xpc_openclose_args *args;
426
427         part_uv->remote_act_state = msg_hdr->act_state;
428
429         switch (msg_hdr->type) {
430         case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
431                 /* syncing of remote_act_state was just done above */
432                 break;
433
434         case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
435                 struct xpc_activate_mq_msg_activate_req_uv *msg;
436
437                 /*
438                  * ??? Do we deal here with ts_jiffies being different
439                  * ??? if act_state != XPC_P_AS_INACTIVE instead of
440                  * ??? below?
441                  */
442                 msg = container_of(msg_hdr, struct
443                                    xpc_activate_mq_msg_activate_req_uv, hdr);
444
445                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
446                 if (part_uv->act_state_req == 0)
447                         xpc_activate_IRQ_rcvd++;
448                 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
449                 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
450                 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
451                 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
452
453                 if (msg->activate_gru_mq_desc_gpa !=
454                     part_uv->activate_gru_mq_desc_gpa) {
455                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
456                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
457                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
458                         part_uv->activate_gru_mq_desc_gpa =
459                             msg->activate_gru_mq_desc_gpa;
460                 }
461                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
462
463                 (*wakeup_hb_checker)++;
464                 break;
465         }
466         case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
467                 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
468
469                 msg = container_of(msg_hdr, struct
470                                    xpc_activate_mq_msg_deactivate_req_uv, hdr);
471
472                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
473                 if (part_uv->act_state_req == 0)
474                         xpc_activate_IRQ_rcvd++;
475                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
476                 part_uv->reason = msg->reason;
477                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
478
479                 (*wakeup_hb_checker)++;
480                 return;
481         }
482         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
483                 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
484
485                 if (!part_setup)
486                         break;
487
488                 msg = container_of(msg_hdr, struct
489                                    xpc_activate_mq_msg_chctl_closerequest_uv,
490                                    hdr);
491                 args = &part->remote_openclose_args[msg->ch_number];
492                 args->reason = msg->reason;
493
494                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
495                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
496                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
497
498                 xpc_wakeup_channel_mgr(part);
499                 break;
500         }
501         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
502                 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
503
504                 if (!part_setup)
505                         break;
506
507                 msg = container_of(msg_hdr, struct
508                                    xpc_activate_mq_msg_chctl_closereply_uv,
509                                    hdr);
510
511                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
512                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
513                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
514
515                 xpc_wakeup_channel_mgr(part);
516                 break;
517         }
518         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
519                 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
520
521                 if (!part_setup)
522                         break;
523
524                 msg = container_of(msg_hdr, struct
525                                    xpc_activate_mq_msg_chctl_openrequest_uv,
526                                    hdr);
527                 args = &part->remote_openclose_args[msg->ch_number];
528                 args->entry_size = msg->entry_size;
529                 args->local_nentries = msg->local_nentries;
530
531                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
532                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
533                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
534
535                 xpc_wakeup_channel_mgr(part);
536                 break;
537         }
538         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
539                 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
540
541                 if (!part_setup)
542                         break;
543
544                 msg = container_of(msg_hdr, struct
545                                    xpc_activate_mq_msg_chctl_openreply_uv, hdr);
546                 args = &part->remote_openclose_args[msg->ch_number];
547                 args->remote_nentries = msg->remote_nentries;
548                 args->local_nentries = msg->local_nentries;
549                 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
550
551                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
552                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
553                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
554
555                 xpc_wakeup_channel_mgr(part);
556                 break;
557         }
558         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
559                 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
560
561                 if (!part_setup)
562                         break;
563
564                 msg = container_of(msg_hdr, struct
565                                 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
566                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
567                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
568                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
569
570                 xpc_wakeup_channel_mgr(part);
571         }
572         case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
573                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
574                 part_uv->flags |= XPC_P_ENGAGED_UV;
575                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
576                 break;
577
578         case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
579                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
580                 part_uv->flags &= ~XPC_P_ENGAGED_UV;
581                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
582                 break;
583
584         default:
585                 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
586                         "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
587
588                 /* get hb checker to deactivate from the remote partition */
589                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
590                 if (part_uv->act_state_req == 0)
591                         xpc_activate_IRQ_rcvd++;
592                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
593                 part_uv->reason = xpBadMsgType;
594                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
595
596                 (*wakeup_hb_checker)++;
597                 return;
598         }
599
600         if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
601             part->remote_rp_ts_jiffies != 0) {
602                 /*
603                  * ??? Does what we do here need to be sensitive to
604                  * ??? act_state or remote_act_state?
605                  */
606                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
607                 if (part_uv->act_state_req == 0)
608                         xpc_activate_IRQ_rcvd++;
609                 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
610                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
611
612                 (*wakeup_hb_checker)++;
613         }
614 }
615
616 static irqreturn_t
617 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
618 {
619         struct xpc_activate_mq_msghdr_uv *msg_hdr;
620         short partid;
621         struct xpc_partition *part;
622         int wakeup_hb_checker = 0;
623         int part_referenced;
624
625         while (1) {
626                 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
627                 if (msg_hdr == NULL)
628                         break;
629
630                 partid = msg_hdr->partid;
631                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
632                         dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
633                                 "received invalid partid=0x%x in message\n",
634                                 partid);
635                 } else {
636                         part = &xpc_partitions[partid];
637
638                         part_referenced = xpc_part_ref(part);
639                         xpc_handle_activate_mq_msg_uv(part, msg_hdr,
640                                                       part_referenced,
641                                                       &wakeup_hb_checker);
642                         if (part_referenced)
643                                 xpc_part_deref(part);
644                 }
645
646                 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
647         }
648
649         if (wakeup_hb_checker)
650                 wake_up_interruptible(&xpc_activate_IRQ_wq);
651
652         return IRQ_HANDLED;
653 }
654
655 static enum xp_retval
656 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
657                                 unsigned long gru_mq_desc_gpa)
658 {
659         enum xp_retval ret;
660
661         ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
662                                sizeof(struct gru_message_queue_desc));
663         if (ret == xpSuccess)
664                 gru_mq_desc->mq = NULL;
665
666         return ret;
667 }
668
669 static enum xp_retval
670 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
671                          int msg_type)
672 {
673         struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
674         struct xpc_partition_uv *part_uv = &part->sn.uv;
675         struct gru_message_queue_desc *gru_mq_desc;
676         unsigned long irq_flags;
677         enum xp_retval ret;
678
679         DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
680
681         msg_hdr->type = msg_type;
682         msg_hdr->partid = xp_partition_id;
683         msg_hdr->act_state = part->act_state;
684         msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
685
686         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
687 again:
688         if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
689                 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
690                 if (gru_mq_desc == NULL) {
691                         gru_mq_desc = kmalloc(sizeof(struct
692                                               gru_message_queue_desc),
693                                               GFP_KERNEL);
694                         if (gru_mq_desc == NULL) {
695                                 ret = xpNoMemory;
696                                 goto done;
697                         }
698                         part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
699                 }
700
701                 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
702                                                       part_uv->
703                                                       activate_gru_mq_desc_gpa);
704                 if (ret != xpSuccess)
705                         goto done;
706
707                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
708                 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
709                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
710         }
711
712         /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
713         ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
714                                msg_size);
715         if (ret != xpSuccess) {
716                 smp_rmb();      /* ensure a fresh copy of part_uv->flags */
717                 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
718                         goto again;
719         }
720 done:
721         mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
722         return ret;
723 }
724
725 static void
726 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
727                               size_t msg_size, int msg_type)
728 {
729         enum xp_retval ret;
730
731         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
732         if (unlikely(ret != xpSuccess))
733                 XPC_DEACTIVATE_PARTITION(part, ret);
734 }
735
736 static void
737 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
738                          void *msg, size_t msg_size, int msg_type)
739 {
740         struct xpc_partition *part = &xpc_partitions[ch->partid];
741         enum xp_retval ret;
742
743         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
744         if (unlikely(ret != xpSuccess)) {
745                 if (irq_flags != NULL)
746                         spin_unlock_irqrestore(&ch->lock, *irq_flags);
747
748                 XPC_DEACTIVATE_PARTITION(part, ret);
749
750                 if (irq_flags != NULL)
751                         spin_lock_irqsave(&ch->lock, *irq_flags);
752         }
753 }
754
755 static void
756 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
757 {
758         unsigned long irq_flags;
759         struct xpc_partition_uv *part_uv = &part->sn.uv;
760
761         /*
762          * !!! Make our side think that the remote partition sent an activate
763          * !!! mq message our way by doing what the activate IRQ handler would
764          * !!! do had one really been sent.
765          */
766
767         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
768         if (part_uv->act_state_req == 0)
769                 xpc_activate_IRQ_rcvd++;
770         part_uv->act_state_req = act_state_req;
771         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
772
773         wake_up_interruptible(&xpc_activate_IRQ_wq);
774 }
775
776 static enum xp_retval
777 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
778                                   size_t *len)
779 {
780         s64 status;
781         enum xp_retval ret;
782
783 #if defined CONFIG_X86_64
784         status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
785                                           (u64 *)len);
786         if (status == BIOS_STATUS_SUCCESS)
787                 ret = xpSuccess;
788         else if (status == BIOS_STATUS_MORE_PASSES)
789                 ret = xpNeedMoreInfo;
790         else
791                 ret = xpBiosError;
792
793 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
794         status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
795         if (status == SALRET_OK)
796                 ret = xpSuccess;
797         else if (status == SALRET_MORE_PASSES)
798                 ret = xpNeedMoreInfo;
799         else
800                 ret = xpSalError;
801
802 #else
803         #error not a supported configuration
804 #endif
805
806         return ret;
807 }
808
809 static int
810 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
811 {
812         xpc_heartbeat_uv =
813             &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
814         rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
815         rp->sn.uv.activate_gru_mq_desc_gpa =
816             uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
817         return 0;
818 }
819
820 static void
821 xpc_allow_hb_uv(short partid)
822 {
823 }
824
825 static void
826 xpc_disallow_hb_uv(short partid)
827 {
828 }
829
830 static void
831 xpc_disallow_all_hbs_uv(void)
832 {
833 }
834
835 static void
836 xpc_increment_heartbeat_uv(void)
837 {
838         xpc_heartbeat_uv->value++;
839 }
840
841 static void
842 xpc_offline_heartbeat_uv(void)
843 {
844         xpc_increment_heartbeat_uv();
845         xpc_heartbeat_uv->offline = 1;
846 }
847
848 static void
849 xpc_online_heartbeat_uv(void)
850 {
851         xpc_increment_heartbeat_uv();
852         xpc_heartbeat_uv->offline = 0;
853 }
854
855 static void
856 xpc_heartbeat_init_uv(void)
857 {
858         xpc_heartbeat_uv->value = 1;
859         xpc_heartbeat_uv->offline = 0;
860 }
861
862 static void
863 xpc_heartbeat_exit_uv(void)
864 {
865         xpc_offline_heartbeat_uv();
866 }
867
868 static enum xp_retval
869 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
870 {
871         struct xpc_partition_uv *part_uv = &part->sn.uv;
872         enum xp_retval ret;
873
874         ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
875                                part_uv->heartbeat_gpa,
876                                sizeof(struct xpc_heartbeat_uv));
877         if (ret != xpSuccess)
878                 return ret;
879
880         if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
881             !part_uv->cached_heartbeat.offline) {
882
883                 ret = xpNoHeartbeat;
884         } else {
885                 part->last_heartbeat = part_uv->cached_heartbeat.value;
886         }
887         return ret;
888 }
889
890 static void
891 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
892                                     unsigned long remote_rp_gpa, int nasid)
893 {
894         short partid = remote_rp->SAL_partid;
895         struct xpc_partition *part = &xpc_partitions[partid];
896         struct xpc_activate_mq_msg_activate_req_uv msg;
897
898         part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
899         part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
900         part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
901         part->sn.uv.activate_gru_mq_desc_gpa =
902             remote_rp->sn.uv.activate_gru_mq_desc_gpa;
903
904         /*
905          * ??? Is it a good idea to make this conditional on what is
906          * ??? potentially stale state information?
907          */
908         if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
909                 msg.rp_gpa = uv_gpa(xpc_rsvd_page);
910                 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
911                 msg.activate_gru_mq_desc_gpa =
912                     xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
913                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
914                                            XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
915         }
916
917         if (part->act_state == XPC_P_AS_INACTIVE)
918                 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
919 }
920
921 static void
922 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
923 {
924         xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
925 }
926
927 static void
928 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
929 {
930         struct xpc_activate_mq_msg_deactivate_req_uv msg;
931
932         /*
933          * ??? Is it a good idea to make this conditional on what is
934          * ??? potentially stale state information?
935          */
936         if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
937             part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
938
939                 msg.reason = part->reason;
940                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
941                                          XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
942         }
943 }
944
945 static void
946 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
947 {
948         /* nothing needs to be done */
949         return;
950 }
951
952 static void
953 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
954 {
955         head->first = NULL;
956         head->last = NULL;
957         spin_lock_init(&head->lock);
958         head->n_entries = 0;
959 }
960
961 static void *
962 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
963 {
964         unsigned long irq_flags;
965         struct xpc_fifo_entry_uv *first;
966
967         spin_lock_irqsave(&head->lock, irq_flags);
968         first = head->first;
969         if (head->first != NULL) {
970                 head->first = first->next;
971                 if (head->first == NULL)
972                         head->last = NULL;
973
974                 head->n_entries--;
975                 BUG_ON(head->n_entries < 0);
976
977                 first->next = NULL;
978         }
979         spin_unlock_irqrestore(&head->lock, irq_flags);
980         return first;
981 }
982
983 static void
984 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
985                       struct xpc_fifo_entry_uv *last)
986 {
987         unsigned long irq_flags;
988
989         last->next = NULL;
990         spin_lock_irqsave(&head->lock, irq_flags);
991         if (head->last != NULL)
992                 head->last->next = last;
993         else
994                 head->first = last;
995         head->last = last;
996         head->n_entries++;
997         spin_unlock_irqrestore(&head->lock, irq_flags);
998 }
999
1000 static int
1001 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
1002 {
1003         return head->n_entries;
1004 }
1005
1006 /*
1007  * Setup the channel structures that are uv specific.
1008  */
1009 static enum xp_retval
1010 xpc_setup_ch_structures_uv(struct xpc_partition *part)
1011 {
1012         struct xpc_channel_uv *ch_uv;
1013         int ch_number;
1014
1015         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1016                 ch_uv = &part->channels[ch_number].sn.uv;
1017
1018                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1019                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1020         }
1021
1022         return xpSuccess;
1023 }
1024
1025 /*
1026  * Teardown the channel structures that are uv specific.
1027  */
1028 static void
1029 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1030 {
1031         /* nothing needs to be done */
1032         return;
1033 }
1034
1035 static enum xp_retval
1036 xpc_make_first_contact_uv(struct xpc_partition *part)
1037 {
1038         struct xpc_activate_mq_msg_uv msg;
1039
1040         /*
1041          * We send a sync msg to get the remote partition's remote_act_state
1042          * updated to our current act_state which at this point should
1043          * be XPC_P_AS_ACTIVATING.
1044          */
1045         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1046                                       XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1047
1048         while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
1049                  (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
1050
1051                 dev_dbg(xpc_part, "waiting to make first contact with "
1052                         "partition %d\n", XPC_PARTID(part));
1053
1054                 /* wait a 1/4 of a second or so */
1055                 (void)msleep_interruptible(250);
1056
1057                 if (part->act_state == XPC_P_AS_DEACTIVATING)
1058                         return part->reason;
1059         }
1060
1061         return xpSuccess;
1062 }
1063
1064 static u64
1065 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1066 {
1067         unsigned long irq_flags;
1068         union xpc_channel_ctl_flags chctl;
1069
1070         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1071         chctl = part->chctl;
1072         if (chctl.all_flags != 0)
1073                 part->chctl.all_flags = 0;
1074
1075         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1076         return chctl.all_flags;
1077 }
1078
1079 static enum xp_retval
1080 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1081 {
1082         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1083         struct xpc_send_msg_slot_uv *msg_slot;
1084         unsigned long irq_flags;
1085         int nentries;
1086         int entry;
1087         size_t nbytes;
1088
1089         for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1090                 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1091                 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1092                 if (ch_uv->send_msg_slots == NULL)
1093                         continue;
1094
1095                 for (entry = 0; entry < nentries; entry++) {
1096                         msg_slot = &ch_uv->send_msg_slots[entry];
1097
1098                         msg_slot->msg_slot_number = entry;
1099                         xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1100                                               &msg_slot->next);
1101                 }
1102
1103                 spin_lock_irqsave(&ch->lock, irq_flags);
1104                 if (nentries < ch->local_nentries)
1105                         ch->local_nentries = nentries;
1106                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1107                 return xpSuccess;
1108         }
1109
1110         return xpNoMemory;
1111 }
1112
1113 static enum xp_retval
1114 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1115 {
1116         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1117         struct xpc_notify_mq_msg_uv *msg_slot;
1118         unsigned long irq_flags;
1119         int nentries;
1120         int entry;
1121         size_t nbytes;
1122
1123         for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1124                 nbytes = nentries * ch->entry_size;
1125                 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1126                 if (ch_uv->recv_msg_slots == NULL)
1127                         continue;
1128
1129                 for (entry = 0; entry < nentries; entry++) {
1130                         msg_slot = ch_uv->recv_msg_slots +
1131                             entry * ch->entry_size;
1132
1133                         msg_slot->hdr.msg_slot_number = entry;
1134                 }
1135
1136                 spin_lock_irqsave(&ch->lock, irq_flags);
1137                 if (nentries < ch->remote_nentries)
1138                         ch->remote_nentries = nentries;
1139                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1140                 return xpSuccess;
1141         }
1142
1143         return xpNoMemory;
1144 }
1145
1146 /*
1147  * Allocate msg_slots associated with the channel.
1148  */
1149 static enum xp_retval
1150 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1151 {
1152         static enum xp_retval ret;
1153         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1154
1155         DBUG_ON(ch->flags & XPC_C_SETUP);
1156
1157         ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1158                                                    gru_message_queue_desc),
1159                                                    GFP_KERNEL);
1160         if (ch_uv->cached_notify_gru_mq_desc == NULL)
1161                 return xpNoMemory;
1162
1163         ret = xpc_allocate_send_msg_slot_uv(ch);
1164         if (ret == xpSuccess) {
1165
1166                 ret = xpc_allocate_recv_msg_slot_uv(ch);
1167                 if (ret != xpSuccess) {
1168                         kfree(ch_uv->send_msg_slots);
1169                         xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1170                 }
1171         }
1172         return ret;
1173 }
1174
1175 /*
1176  * Free up msg_slots and clear other stuff that were setup for the specified
1177  * channel.
1178  */
1179 static void
1180 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1181 {
1182         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1183
1184         DBUG_ON(!spin_is_locked(&ch->lock));
1185
1186         kfree(ch_uv->cached_notify_gru_mq_desc);
1187         ch_uv->cached_notify_gru_mq_desc = NULL;
1188
1189         if (ch->flags & XPC_C_SETUP) {
1190                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1191                 kfree(ch_uv->send_msg_slots);
1192                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1193                 kfree(ch_uv->recv_msg_slots);
1194         }
1195 }
1196
1197 static void
1198 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1199 {
1200         struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1201
1202         msg.ch_number = ch->number;
1203         msg.reason = ch->reason;
1204         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1205                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1206 }
1207
1208 static void
1209 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1210 {
1211         struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1212
1213         msg.ch_number = ch->number;
1214         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1215                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1216 }
1217
1218 static void
1219 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1220 {
1221         struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1222
1223         msg.ch_number = ch->number;
1224         msg.entry_size = ch->entry_size;
1225         msg.local_nentries = ch->local_nentries;
1226         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1227                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1228 }
1229
1230 static void
1231 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1232 {
1233         struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1234
1235         msg.ch_number = ch->number;
1236         msg.local_nentries = ch->local_nentries;
1237         msg.remote_nentries = ch->remote_nentries;
1238         msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1239         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1240                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1241 }
1242
1243 static void
1244 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1245 {
1246         struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1247
1248         msg.ch_number = ch->number;
1249         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1250                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1251 }
1252
1253 static void
1254 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1255 {
1256         unsigned long irq_flags;
1257
1258         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1259         part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1260         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1261
1262         xpc_wakeup_channel_mgr(part);
1263 }
1264
1265 static enum xp_retval
1266 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1267                                unsigned long gru_mq_desc_gpa)
1268 {
1269         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1270
1271         DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1272         return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1273                                                gru_mq_desc_gpa);
1274 }
1275
1276 static void
1277 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1278 {
1279         struct xpc_activate_mq_msg_uv msg;
1280
1281         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1282                                       XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1283 }
1284
1285 static void
1286 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1287 {
1288         struct xpc_activate_mq_msg_uv msg;
1289
1290         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1291                                       XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1292 }
1293
1294 static void
1295 xpc_assume_partition_disengaged_uv(short partid)
1296 {
1297         struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1298         unsigned long irq_flags;
1299
1300         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1301         part_uv->flags &= ~XPC_P_ENGAGED_UV;
1302         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1303 }
1304
1305 static int
1306 xpc_partition_engaged_uv(short partid)
1307 {
1308         return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1309 }
1310
1311 static int
1312 xpc_any_partition_engaged_uv(void)
1313 {
1314         struct xpc_partition_uv *part_uv;
1315         short partid;
1316
1317         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1318                 part_uv = &xpc_partitions[partid].sn.uv;
1319                 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1320                         return 1;
1321         }
1322         return 0;
1323 }
1324
1325 static enum xp_retval
1326 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1327                          struct xpc_send_msg_slot_uv **address_of_msg_slot)
1328 {
1329         enum xp_retval ret;
1330         struct xpc_send_msg_slot_uv *msg_slot;
1331         struct xpc_fifo_entry_uv *entry;
1332
1333         while (1) {
1334                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1335                 if (entry != NULL)
1336                         break;
1337
1338                 if (flags & XPC_NOWAIT)
1339                         return xpNoWait;
1340
1341                 ret = xpc_allocate_msg_wait(ch);
1342                 if (ret != xpInterrupted && ret != xpTimeout)
1343                         return ret;
1344         }
1345
1346         msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1347         *address_of_msg_slot = msg_slot;
1348         return xpSuccess;
1349 }
1350
1351 static void
1352 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1353                      struct xpc_send_msg_slot_uv *msg_slot)
1354 {
1355         xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1356
1357         /* wakeup anyone waiting for a free msg slot */
1358         if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1359                 wake_up(&ch->msg_allocate_wq);
1360 }
1361
1362 static void
1363 xpc_notify_sender_uv(struct xpc_channel *ch,
1364                      struct xpc_send_msg_slot_uv *msg_slot,
1365                      enum xp_retval reason)
1366 {
1367         xpc_notify_func func = msg_slot->func;
1368
1369         if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1370
1371                 atomic_dec(&ch->n_to_notify);
1372
1373                 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1374                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1375                         msg_slot->msg_slot_number, ch->partid, ch->number);
1376
1377                 func(reason, ch->partid, ch->number, msg_slot->key);
1378
1379                 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1380                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1381                         msg_slot->msg_slot_number, ch->partid, ch->number);
1382         }
1383 }
1384
1385 static void
1386 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1387                             struct xpc_notify_mq_msg_uv *msg)
1388 {
1389         struct xpc_send_msg_slot_uv *msg_slot;
1390         int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1391
1392         msg_slot = &ch->sn.uv.send_msg_slots[entry];
1393
1394         BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1395         msg_slot->msg_slot_number += ch->local_nentries;
1396
1397         if (msg_slot->func != NULL)
1398                 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1399
1400         xpc_free_msg_slot_uv(ch, msg_slot);
1401 }
1402
1403 static void
1404 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1405                             struct xpc_notify_mq_msg_uv *msg)
1406 {
1407         struct xpc_partition_uv *part_uv = &part->sn.uv;
1408         struct xpc_channel *ch;
1409         struct xpc_channel_uv *ch_uv;
1410         struct xpc_notify_mq_msg_uv *msg_slot;
1411         unsigned long irq_flags;
1412         int ch_number = msg->hdr.ch_number;
1413
1414         if (unlikely(ch_number >= part->nchannels)) {
1415                 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1416                         "channel number=0x%x in message from partid=%d\n",
1417                         ch_number, XPC_PARTID(part));
1418
1419                 /* get hb checker to deactivate from the remote partition */
1420                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1421                 if (part_uv->act_state_req == 0)
1422                         xpc_activate_IRQ_rcvd++;
1423                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1424                 part_uv->reason = xpBadChannelNumber;
1425                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1426
1427                 wake_up_interruptible(&xpc_activate_IRQ_wq);
1428                 return;
1429         }
1430
1431         ch = &part->channels[ch_number];
1432         xpc_msgqueue_ref(ch);
1433
1434         if (!(ch->flags & XPC_C_CONNECTED)) {
1435                 xpc_msgqueue_deref(ch);
1436                 return;
1437         }
1438
1439         /* see if we're really dealing with an ACK for a previously sent msg */
1440         if (msg->hdr.size == 0) {
1441                 xpc_handle_notify_mq_ack_uv(ch, msg);
1442                 xpc_msgqueue_deref(ch);
1443                 return;
1444         }
1445
1446         /* we're dealing with a normal message sent via the notify_mq */
1447         ch_uv = &ch->sn.uv;
1448
1449         msg_slot = ch_uv->recv_msg_slots +
1450             (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1451
1452         BUG_ON(msg_slot->hdr.size != 0);
1453
1454         memcpy(msg_slot, msg, msg->hdr.size);
1455
1456         xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1457
1458         if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1459                 /*
1460                  * If there is an existing idle kthread get it to deliver
1461                  * the payload, otherwise we'll have to get the channel mgr
1462                  * for this partition to create a kthread to do the delivery.
1463                  */
1464                 if (atomic_read(&ch->kthreads_idle) > 0)
1465                         wake_up_nr(&ch->idle_wq, 1);
1466                 else
1467                         xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1468         }
1469         xpc_msgqueue_deref(ch);
1470 }
1471
1472 static irqreturn_t
1473 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1474 {
1475         struct xpc_notify_mq_msg_uv *msg;
1476         short partid;
1477         struct xpc_partition *part;
1478
1479         while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1480                NULL) {
1481
1482                 partid = msg->hdr.partid;
1483                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1484                         dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1485                                 "invalid partid=0x%x in message\n", partid);
1486                 } else {
1487                         part = &xpc_partitions[partid];
1488
1489                         if (xpc_part_ref(part)) {
1490                                 xpc_handle_notify_mq_msg_uv(part, msg);
1491                                 xpc_part_deref(part);
1492                         }
1493                 }
1494
1495                 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1496         }
1497
1498         return IRQ_HANDLED;
1499 }
1500
1501 static int
1502 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1503 {
1504         return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1505 }
1506
1507 static void
1508 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1509 {
1510         struct xpc_channel *ch = &part->channels[ch_number];
1511         int ndeliverable_payloads;
1512
1513         xpc_msgqueue_ref(ch);
1514
1515         ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1516
1517         if (ndeliverable_payloads > 0 &&
1518             (ch->flags & XPC_C_CONNECTED) &&
1519             (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1520
1521                 xpc_activate_kthreads(ch, ndeliverable_payloads);
1522         }
1523
1524         xpc_msgqueue_deref(ch);
1525 }
1526
1527 static enum xp_retval
1528 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1529                     u16 payload_size, u8 notify_type, xpc_notify_func func,
1530                     void *key)
1531 {
1532         enum xp_retval ret = xpSuccess;
1533         struct xpc_send_msg_slot_uv *msg_slot = NULL;
1534         struct xpc_notify_mq_msg_uv *msg;
1535         u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1536         size_t msg_size;
1537
1538         DBUG_ON(notify_type != XPC_N_CALL);
1539
1540         msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1541         if (msg_size > ch->entry_size)
1542                 return xpPayloadTooBig;
1543
1544         xpc_msgqueue_ref(ch);
1545
1546         if (ch->flags & XPC_C_DISCONNECTING) {
1547                 ret = ch->reason;
1548                 goto out_1;
1549         }
1550         if (!(ch->flags & XPC_C_CONNECTED)) {
1551                 ret = xpNotConnected;
1552                 goto out_1;
1553         }
1554
1555         ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1556         if (ret != xpSuccess)
1557                 goto out_1;
1558
1559         if (func != NULL) {
1560                 atomic_inc(&ch->n_to_notify);
1561
1562                 msg_slot->key = key;
1563                 smp_wmb(); /* a non-NULL func must hit memory after the key */
1564                 msg_slot->func = func;
1565
1566                 if (ch->flags & XPC_C_DISCONNECTING) {
1567                         ret = ch->reason;
1568                         goto out_2;
1569                 }
1570         }
1571
1572         msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1573         msg->hdr.partid = xp_partition_id;
1574         msg->hdr.ch_number = ch->number;
1575         msg->hdr.size = msg_size;
1576         msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1577         memcpy(&msg->payload, payload, payload_size);
1578
1579         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1580                                msg_size);
1581         if (ret == xpSuccess)
1582                 goto out_1;
1583
1584         XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1585 out_2:
1586         if (func != NULL) {
1587                 /*
1588                  * Try to NULL the msg_slot's func field. If we fail, then
1589                  * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1590                  * case we need to pretend we succeeded to send the message
1591                  * since the user will get a callout for the disconnect error
1592                  * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1593                  * error returned here will confuse them. Additionally, since
1594                  * in this case the channel is being disconnected we don't need
1595                  * to put the the msg_slot back on the free list.
1596                  */
1597                 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1598                         ret = xpSuccess;
1599                         goto out_1;
1600                 }
1601
1602                 msg_slot->key = NULL;
1603                 atomic_dec(&ch->n_to_notify);
1604         }
1605         xpc_free_msg_slot_uv(ch, msg_slot);
1606 out_1:
1607         xpc_msgqueue_deref(ch);
1608         return ret;
1609 }
1610
1611 /*
1612  * Tell the callers of xpc_send_notify() that the status of their payloads
1613  * is unknown because the channel is now disconnecting.
1614  *
1615  * We don't worry about putting these msg_slots on the free list since the
1616  * msg_slots themselves are about to be kfree'd.
1617  */
1618 static void
1619 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1620 {
1621         struct xpc_send_msg_slot_uv *msg_slot;
1622         int entry;
1623
1624         DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1625
1626         for (entry = 0; entry < ch->local_nentries; entry++) {
1627
1628                 if (atomic_read(&ch->n_to_notify) == 0)
1629                         break;
1630
1631                 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1632                 if (msg_slot->func != NULL)
1633                         xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1634         }
1635 }
1636
1637 /*
1638  * Get the next deliverable message's payload.
1639  */
1640 static void *
1641 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1642 {
1643         struct xpc_fifo_entry_uv *entry;
1644         struct xpc_notify_mq_msg_uv *msg;
1645         void *payload = NULL;
1646
1647         if (!(ch->flags & XPC_C_DISCONNECTING)) {
1648                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1649                 if (entry != NULL) {
1650                         msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1651                                            hdr.u.next);
1652                         payload = &msg->payload;
1653                 }
1654         }
1655         return payload;
1656 }
1657
1658 static void
1659 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1660 {
1661         struct xpc_notify_mq_msg_uv *msg;
1662         enum xp_retval ret;
1663
1664         msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1665
1666         /* return an ACK to the sender of this message */
1667
1668         msg->hdr.partid = xp_partition_id;
1669         msg->hdr.size = 0;      /* size of zero indicates this is an ACK */
1670
1671         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1672                                sizeof(struct xpc_notify_mq_msghdr_uv));
1673         if (ret != xpSuccess)
1674                 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1675 }
1676
1677 static struct xpc_arch_operations xpc_arch_ops_uv = {
1678         .setup_partitions = xpc_setup_partitions_uv,
1679         .teardown_partitions = xpc_teardown_partitions_uv,
1680         .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1681         .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1682         .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1683
1684         .allow_hb = xpc_allow_hb_uv,
1685         .disallow_hb = xpc_disallow_hb_uv,
1686         .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1687         .increment_heartbeat = xpc_increment_heartbeat_uv,
1688         .offline_heartbeat = xpc_offline_heartbeat_uv,
1689         .online_heartbeat = xpc_online_heartbeat_uv,
1690         .heartbeat_init = xpc_heartbeat_init_uv,
1691         .heartbeat_exit = xpc_heartbeat_exit_uv,
1692         .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1693
1694         .request_partition_activation =
1695                 xpc_request_partition_activation_uv,
1696         .request_partition_reactivation =
1697                 xpc_request_partition_reactivation_uv,
1698         .request_partition_deactivation =
1699                 xpc_request_partition_deactivation_uv,
1700         .cancel_partition_deactivation_request =
1701                 xpc_cancel_partition_deactivation_request_uv,
1702
1703         .setup_ch_structures = xpc_setup_ch_structures_uv,
1704         .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1705
1706         .make_first_contact = xpc_make_first_contact_uv,
1707
1708         .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1709         .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1710         .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1711         .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1712         .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1713         .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1714         .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1715
1716         .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1717
1718         .setup_msg_structures = xpc_setup_msg_structures_uv,
1719         .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1720
1721         .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1722         .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1723         .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1724         .partition_engaged = xpc_partition_engaged_uv,
1725         .any_partition_engaged = xpc_any_partition_engaged_uv,
1726
1727         .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1728         .send_payload = xpc_send_payload_uv,
1729         .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1730         .received_payload = xpc_received_payload_uv,
1731         .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1732 };
1733
1734 int
1735 xpc_init_uv(void)
1736 {
1737         xpc_arch_ops = xpc_arch_ops_uv;
1738
1739         if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1740                 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1741                         XPC_MSG_HDR_MAX_SIZE);
1742                 return -E2BIG;
1743         }
1744
1745         xpc_activate_mq_uv = xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, 0,
1746                                                   XPC_ACTIVATE_IRQ_NAME,
1747                                                   xpc_handle_activate_IRQ_uv);
1748         if (IS_ERR(xpc_activate_mq_uv))
1749                 return PTR_ERR(xpc_activate_mq_uv);
1750
1751         xpc_notify_mq_uv = xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, 0,
1752                                                 XPC_NOTIFY_IRQ_NAME,
1753                                                 xpc_handle_notify_IRQ_uv);
1754         if (IS_ERR(xpc_notify_mq_uv)) {
1755                 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1756                 return PTR_ERR(xpc_notify_mq_uv);
1757         }
1758
1759         return 0;
1760 }
1761
1762 void
1763 xpc_exit_uv(void)
1764 {
1765         xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1766         xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1767 }