page allocator: do not check NUMA node ID when the caller knows the node is valid
[sfrench/cifs-2.6.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 <asm/uv/uv_hub.h>
23 #if defined CONFIG_X86_64
24 #include <asm/uv/bios.h>
25 #include <asm/uv/uv_irq.h>
26 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
27 #include <asm/sn/intr.h>
28 #include <asm/sn/sn_sal.h>
29 #endif
30 #include "../sgi-gru/gru.h"
31 #include "../sgi-gru/grukservices.h"
32 #include "xpc.h"
33
34 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
35 struct uv_IO_APIC_route_entry {
36         __u64   vector          :  8,
37                 delivery_mode   :  3,
38                 dest_mode       :  1,
39                 delivery_status :  1,
40                 polarity        :  1,
41                 __reserved_1    :  1,
42                 trigger         :  1,
43                 mask            :  1,
44                 __reserved_2    : 15,
45                 dest            : 32;
46 };
47 #endif
48
49 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
50
51 #define XPC_ACTIVATE_MSG_SIZE_UV        (1 * GRU_CACHE_LINE_BYTES)
52 #define XPC_ACTIVATE_MQ_SIZE_UV         (4 * XP_MAX_NPARTITIONS_UV * \
53                                          XPC_ACTIVATE_MSG_SIZE_UV)
54 #define XPC_ACTIVATE_IRQ_NAME           "xpc_activate"
55
56 #define XPC_NOTIFY_MSG_SIZE_UV          (2 * GRU_CACHE_LINE_BYTES)
57 #define XPC_NOTIFY_MQ_SIZE_UV           (4 * XP_MAX_NPARTITIONS_UV * \
58                                          XPC_NOTIFY_MSG_SIZE_UV)
59 #define XPC_NOTIFY_IRQ_NAME             "xpc_notify"
60
61 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
62 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
63
64 static int
65 xpc_setup_partitions_uv(void)
66 {
67         short partid;
68         struct xpc_partition_uv *part_uv;
69
70         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
71                 part_uv = &xpc_partitions[partid].sn.uv;
72
73                 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
74                 spin_lock_init(&part_uv->flags_lock);
75                 part_uv->remote_act_state = XPC_P_AS_INACTIVE;
76         }
77         return 0;
78 }
79
80 static void
81 xpc_teardown_partitions_uv(void)
82 {
83         short partid;
84         struct xpc_partition_uv *part_uv;
85         unsigned long irq_flags;
86
87         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
88                 part_uv = &xpc_partitions[partid].sn.uv;
89
90                 if (part_uv->cached_activate_gru_mq_desc != NULL) {
91                         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
92                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
93                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
94                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
95                         kfree(part_uv->cached_activate_gru_mq_desc);
96                         part_uv->cached_activate_gru_mq_desc = NULL;
97                         mutex_unlock(&part_uv->
98                                      cached_activate_gru_mq_desc_mutex);
99                 }
100         }
101 }
102
103 static int
104 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
105 {
106         int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
107
108 #if defined CONFIG_X86_64
109         mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset);
110         if (mq->irq < 0) {
111                 dev_err(xpc_part, "uv_setup_irq() returned error=%d\n",
112                         -mq->irq);
113                 return mq->irq;
114         }
115
116         mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
117
118 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
119         if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
120                 mq->irq = SGI_XPC_ACTIVATE;
121         else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
122                 mq->irq = SGI_XPC_NOTIFY;
123         else
124                 return -EINVAL;
125
126         mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
127         uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
128 #else
129         #error not a supported configuration
130 #endif
131
132         return 0;
133 }
134
135 static void
136 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
137 {
138 #if defined CONFIG_X86_64
139         uv_teardown_irq(mq->irq, mq->mmr_blade, mq->mmr_offset);
140
141 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
142         int mmr_pnode;
143         unsigned long mmr_value;
144
145         mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
146         mmr_value = 1UL << 16;
147
148         uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
149 #else
150         #error not a supported configuration
151 #endif
152 }
153
154 static int
155 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
156 {
157         int ret;
158
159 #if defined CONFIG_X86_64
160         ret = uv_bios_mq_watchlist_alloc(mq->mmr_blade, uv_gpa(mq->address),
161                                          mq->order, &mq->mmr_offset);
162         if (ret < 0) {
163                 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
164                         "ret=%d\n", ret);
165                 return ret;
166         }
167 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
168         ret = sn_mq_watchlist_alloc(mq->mmr_blade, (void *)uv_gpa(mq->address),
169                                     mq->order, &mq->mmr_offset);
170         if (ret < 0) {
171                 dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
172                         ret);
173                 return -EBUSY;
174         }
175 #else
176         #error not a supported configuration
177 #endif
178
179         mq->watchlist_num = ret;
180         return 0;
181 }
182
183 static void
184 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
185 {
186         int ret;
187
188 #if defined CONFIG_X86_64
189         ret = uv_bios_mq_watchlist_free(mq->mmr_blade, mq->watchlist_num);
190         BUG_ON(ret != BIOS_STATUS_SUCCESS);
191 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
192         ret = sn_mq_watchlist_free(mq->mmr_blade, mq->watchlist_num);
193         BUG_ON(ret != SALRET_OK);
194 #else
195         #error not a supported configuration
196 #endif
197 }
198
199 static struct xpc_gru_mq_uv *
200 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
201                      irq_handler_t irq_handler)
202 {
203         enum xp_retval xp_ret;
204         int ret;
205         int nid;
206         int pg_order;
207         struct page *page;
208         struct xpc_gru_mq_uv *mq;
209         struct uv_IO_APIC_route_entry *mmr_value;
210
211         mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
212         if (mq == NULL) {
213                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
214                         "a xpc_gru_mq_uv structure\n");
215                 ret = -ENOMEM;
216                 goto out_0;
217         }
218
219         mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
220                                   GFP_KERNEL);
221         if (mq->gru_mq_desc == NULL) {
222                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
223                         "a gru_message_queue_desc structure\n");
224                 ret = -ENOMEM;
225                 goto out_1;
226         }
227
228         pg_order = get_order(mq_size);
229         mq->order = pg_order + PAGE_SHIFT;
230         mq_size = 1UL << mq->order;
231
232         mq->mmr_blade = uv_cpu_to_blade_id(cpu);
233
234         nid = cpu_to_node(cpu);
235         page = alloc_pages_exact_node(nid, GFP_KERNEL | __GFP_ZERO | GFP_THISNODE,
236                                 pg_order);
237         if (page == NULL) {
238                 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
239                         "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
240                 ret = -ENOMEM;
241                 goto out_2;
242         }
243         mq->address = page_address(page);
244
245         /* enable generation of irq when GRU mq operation occurs to this mq */
246         ret = xpc_gru_mq_watchlist_alloc_uv(mq);
247         if (ret != 0)
248                 goto out_3;
249
250         ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
251         if (ret != 0)
252                 goto out_4;
253
254         ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
255         if (ret != 0) {
256                 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
257                         mq->irq, -ret);
258                 goto out_5;
259         }
260
261         mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
262         ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
263                                        nid, mmr_value->vector, mmr_value->dest);
264         if (ret != 0) {
265                 dev_err(xpc_part, "gru_create_message_queue() returned "
266                         "error=%d\n", ret);
267                 ret = -EINVAL;
268                 goto out_6;
269         }
270
271         /* allow other partitions to access this GRU mq */
272         xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
273         if (xp_ret != xpSuccess) {
274                 ret = -EACCES;
275                 goto out_6;
276         }
277
278         return mq;
279
280         /* something went wrong */
281 out_6:
282         free_irq(mq->irq, NULL);
283 out_5:
284         xpc_release_gru_mq_irq_uv(mq);
285 out_4:
286         xpc_gru_mq_watchlist_free_uv(mq);
287 out_3:
288         free_pages((unsigned long)mq->address, pg_order);
289 out_2:
290         kfree(mq->gru_mq_desc);
291 out_1:
292         kfree(mq);
293 out_0:
294         return ERR_PTR(ret);
295 }
296
297 static void
298 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
299 {
300         unsigned int mq_size;
301         int pg_order;
302         int ret;
303
304         /* disallow other partitions to access GRU mq */
305         mq_size = 1UL << mq->order;
306         ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
307         BUG_ON(ret != xpSuccess);
308
309         /* unregister irq handler and release mq irq/vector mapping */
310         free_irq(mq->irq, NULL);
311         xpc_release_gru_mq_irq_uv(mq);
312
313         /* disable generation of irq when GRU mq op occurs to this mq */
314         xpc_gru_mq_watchlist_free_uv(mq);
315
316         pg_order = mq->order - PAGE_SHIFT;
317         free_pages((unsigned long)mq->address, pg_order);
318
319         kfree(mq);
320 }
321
322 static enum xp_retval
323 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
324                  size_t msg_size)
325 {
326         enum xp_retval xp_ret;
327         int ret;
328
329         while (1) {
330                 ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
331                 if (ret == MQE_OK) {
332                         xp_ret = xpSuccess;
333                         break;
334                 }
335
336                 if (ret == MQE_QUEUE_FULL) {
337                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
338                                 "error=MQE_QUEUE_FULL\n");
339                         /* !!! handle QLimit reached; delay & try again */
340                         /* ??? Do we add a limit to the number of retries? */
341                         (void)msleep_interruptible(10);
342                 } else if (ret == MQE_CONGESTION) {
343                         dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
344                                 "error=MQE_CONGESTION\n");
345                         /* !!! handle LB Overflow; simply try again */
346                         /* ??? Do we add a limit to the number of retries? */
347                 } else {
348                         /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
349                         dev_err(xpc_chan, "gru_send_message_gpa() returned "
350                                 "error=%d\n", ret);
351                         xp_ret = xpGruSendMqError;
352                         break;
353                 }
354         }
355         return xp_ret;
356 }
357
358 static void
359 xpc_process_activate_IRQ_rcvd_uv(void)
360 {
361         unsigned long irq_flags;
362         short partid;
363         struct xpc_partition *part;
364         u8 act_state_req;
365
366         DBUG_ON(xpc_activate_IRQ_rcvd == 0);
367
368         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
369         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
370                 part = &xpc_partitions[partid];
371
372                 if (part->sn.uv.act_state_req == 0)
373                         continue;
374
375                 xpc_activate_IRQ_rcvd--;
376                 BUG_ON(xpc_activate_IRQ_rcvd < 0);
377
378                 act_state_req = part->sn.uv.act_state_req;
379                 part->sn.uv.act_state_req = 0;
380                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
381
382                 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
383                         if (part->act_state == XPC_P_AS_INACTIVE)
384                                 xpc_activate_partition(part);
385                         else if (part->act_state == XPC_P_AS_DEACTIVATING)
386                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
387
388                 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
389                         if (part->act_state == XPC_P_AS_INACTIVE)
390                                 xpc_activate_partition(part);
391                         else
392                                 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
393
394                 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
395                         XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
396
397                 } else {
398                         BUG();
399                 }
400
401                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
402                 if (xpc_activate_IRQ_rcvd == 0)
403                         break;
404         }
405         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
406
407 }
408
409 static void
410 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
411                               struct xpc_activate_mq_msghdr_uv *msg_hdr,
412                               int *wakeup_hb_checker)
413 {
414         unsigned long irq_flags;
415         struct xpc_partition_uv *part_uv = &part->sn.uv;
416         struct xpc_openclose_args *args;
417
418         part_uv->remote_act_state = msg_hdr->act_state;
419
420         switch (msg_hdr->type) {
421         case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
422                 /* syncing of remote_act_state was just done above */
423                 break;
424
425         case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
426                 struct xpc_activate_mq_msg_activate_req_uv *msg;
427
428                 /*
429                  * ??? Do we deal here with ts_jiffies being different
430                  * ??? if act_state != XPC_P_AS_INACTIVE instead of
431                  * ??? below?
432                  */
433                 msg = container_of(msg_hdr, struct
434                                    xpc_activate_mq_msg_activate_req_uv, hdr);
435
436                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
437                 if (part_uv->act_state_req == 0)
438                         xpc_activate_IRQ_rcvd++;
439                 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
440                 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
441                 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
442                 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
443
444                 if (msg->activate_gru_mq_desc_gpa !=
445                     part_uv->activate_gru_mq_desc_gpa) {
446                         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
447                         part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
448                         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
449                         part_uv->activate_gru_mq_desc_gpa =
450                             msg->activate_gru_mq_desc_gpa;
451                 }
452                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
453
454                 (*wakeup_hb_checker)++;
455                 break;
456         }
457         case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
458                 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
459
460                 msg = container_of(msg_hdr, struct
461                                    xpc_activate_mq_msg_deactivate_req_uv, hdr);
462
463                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
464                 if (part_uv->act_state_req == 0)
465                         xpc_activate_IRQ_rcvd++;
466                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
467                 part_uv->reason = msg->reason;
468                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
469
470                 (*wakeup_hb_checker)++;
471                 return;
472         }
473         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
474                 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
475
476                 msg = container_of(msg_hdr, struct
477                                    xpc_activate_mq_msg_chctl_closerequest_uv,
478                                    hdr);
479                 args = &part->remote_openclose_args[msg->ch_number];
480                 args->reason = msg->reason;
481
482                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
483                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
484                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
485
486                 xpc_wakeup_channel_mgr(part);
487                 break;
488         }
489         case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
490                 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
491
492                 msg = container_of(msg_hdr, struct
493                                    xpc_activate_mq_msg_chctl_closereply_uv,
494                                    hdr);
495
496                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
497                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
498                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
499
500                 xpc_wakeup_channel_mgr(part);
501                 break;
502         }
503         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
504                 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
505
506                 msg = container_of(msg_hdr, struct
507                                    xpc_activate_mq_msg_chctl_openrequest_uv,
508                                    hdr);
509                 args = &part->remote_openclose_args[msg->ch_number];
510                 args->entry_size = msg->entry_size;
511                 args->local_nentries = msg->local_nentries;
512
513                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
514                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
515                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
516
517                 xpc_wakeup_channel_mgr(part);
518                 break;
519         }
520         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
521                 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
522
523                 msg = container_of(msg_hdr, struct
524                                    xpc_activate_mq_msg_chctl_openreply_uv, hdr);
525                 args = &part->remote_openclose_args[msg->ch_number];
526                 args->remote_nentries = msg->remote_nentries;
527                 args->local_nentries = msg->local_nentries;
528                 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
529
530                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
531                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
532                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
533
534                 xpc_wakeup_channel_mgr(part);
535                 break;
536         }
537         case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
538                 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
539
540                 msg = container_of(msg_hdr, struct
541                                 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
542                 spin_lock_irqsave(&part->chctl_lock, irq_flags);
543                 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
544                 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
545
546                 xpc_wakeup_channel_mgr(part);
547         }
548         case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
549                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
550                 part_uv->flags |= XPC_P_ENGAGED_UV;
551                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
552                 break;
553
554         case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
555                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
556                 part_uv->flags &= ~XPC_P_ENGAGED_UV;
557                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
558                 break;
559
560         default:
561                 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
562                         "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
563
564                 /* get hb checker to deactivate from the remote partition */
565                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
566                 if (part_uv->act_state_req == 0)
567                         xpc_activate_IRQ_rcvd++;
568                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
569                 part_uv->reason = xpBadMsgType;
570                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
571
572                 (*wakeup_hb_checker)++;
573                 return;
574         }
575
576         if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
577             part->remote_rp_ts_jiffies != 0) {
578                 /*
579                  * ??? Does what we do here need to be sensitive to
580                  * ??? act_state or remote_act_state?
581                  */
582                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
583                 if (part_uv->act_state_req == 0)
584                         xpc_activate_IRQ_rcvd++;
585                 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
586                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
587
588                 (*wakeup_hb_checker)++;
589         }
590 }
591
592 static irqreturn_t
593 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
594 {
595         struct xpc_activate_mq_msghdr_uv *msg_hdr;
596         short partid;
597         struct xpc_partition *part;
598         int wakeup_hb_checker = 0;
599         int part_referenced;
600
601         while (1) {
602                 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
603                 if (msg_hdr == NULL)
604                         break;
605
606                 partid = msg_hdr->partid;
607                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
608                         dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
609                                 "received invalid partid=0x%x in message\n",
610                                 partid);
611                 } else {
612                         part = &xpc_partitions[partid];
613
614                         part_referenced = xpc_part_ref(part);
615                         xpc_handle_activate_mq_msg_uv(part, msg_hdr,
616                                                       &wakeup_hb_checker);
617                         if (part_referenced)
618                                 xpc_part_deref(part);
619                 }
620
621                 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
622         }
623
624         if (wakeup_hb_checker)
625                 wake_up_interruptible(&xpc_activate_IRQ_wq);
626
627         return IRQ_HANDLED;
628 }
629
630 static enum xp_retval
631 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
632                                 unsigned long gru_mq_desc_gpa)
633 {
634         enum xp_retval ret;
635
636         ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
637                                sizeof(struct gru_message_queue_desc));
638         if (ret == xpSuccess)
639                 gru_mq_desc->mq = NULL;
640
641         return ret;
642 }
643
644 static enum xp_retval
645 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
646                          int msg_type)
647 {
648         struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
649         struct xpc_partition_uv *part_uv = &part->sn.uv;
650         struct gru_message_queue_desc *gru_mq_desc;
651         unsigned long irq_flags;
652         enum xp_retval ret;
653
654         DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
655
656         msg_hdr->type = msg_type;
657         msg_hdr->partid = xp_partition_id;
658         msg_hdr->act_state = part->act_state;
659         msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
660
661         mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
662 again:
663         if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
664                 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
665                 if (gru_mq_desc == NULL) {
666                         gru_mq_desc = kmalloc(sizeof(struct
667                                               gru_message_queue_desc),
668                                               GFP_KERNEL);
669                         if (gru_mq_desc == NULL) {
670                                 ret = xpNoMemory;
671                                 goto done;
672                         }
673                         part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
674                 }
675
676                 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
677                                                       part_uv->
678                                                       activate_gru_mq_desc_gpa);
679                 if (ret != xpSuccess)
680                         goto done;
681
682                 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
683                 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
684                 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
685         }
686
687         /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
688         ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
689                                msg_size);
690         if (ret != xpSuccess) {
691                 smp_rmb();      /* ensure a fresh copy of part_uv->flags */
692                 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
693                         goto again;
694         }
695 done:
696         mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
697         return ret;
698 }
699
700 static void
701 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
702                               size_t msg_size, int msg_type)
703 {
704         enum xp_retval ret;
705
706         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
707         if (unlikely(ret != xpSuccess))
708                 XPC_DEACTIVATE_PARTITION(part, ret);
709 }
710
711 static void
712 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
713                          void *msg, size_t msg_size, int msg_type)
714 {
715         struct xpc_partition *part = &xpc_partitions[ch->partid];
716         enum xp_retval ret;
717
718         ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
719         if (unlikely(ret != xpSuccess)) {
720                 if (irq_flags != NULL)
721                         spin_unlock_irqrestore(&ch->lock, *irq_flags);
722
723                 XPC_DEACTIVATE_PARTITION(part, ret);
724
725                 if (irq_flags != NULL)
726                         spin_lock_irqsave(&ch->lock, *irq_flags);
727         }
728 }
729
730 static void
731 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
732 {
733         unsigned long irq_flags;
734         struct xpc_partition_uv *part_uv = &part->sn.uv;
735
736         /*
737          * !!! Make our side think that the remote partition sent an activate
738          * !!! mq message our way by doing what the activate IRQ handler would
739          * !!! do had one really been sent.
740          */
741
742         spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
743         if (part_uv->act_state_req == 0)
744                 xpc_activate_IRQ_rcvd++;
745         part_uv->act_state_req = act_state_req;
746         spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
747
748         wake_up_interruptible(&xpc_activate_IRQ_wq);
749 }
750
751 static enum xp_retval
752 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
753                                   size_t *len)
754 {
755         s64 status;
756         enum xp_retval ret;
757
758 #if defined CONFIG_X86_64
759         status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
760                                           (u64 *)len);
761         if (status == BIOS_STATUS_SUCCESS)
762                 ret = xpSuccess;
763         else if (status == BIOS_STATUS_MORE_PASSES)
764                 ret = xpNeedMoreInfo;
765         else
766                 ret = xpBiosError;
767
768 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
769         status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
770         if (status == SALRET_OK)
771                 ret = xpSuccess;
772         else if (status == SALRET_MORE_PASSES)
773                 ret = xpNeedMoreInfo;
774         else
775                 ret = xpSalError;
776
777 #else
778         #error not a supported configuration
779 #endif
780
781         return ret;
782 }
783
784 static int
785 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
786 {
787         xpc_heartbeat_uv =
788             &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
789         rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
790         rp->sn.uv.activate_gru_mq_desc_gpa =
791             uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
792         return 0;
793 }
794
795 static void
796 xpc_allow_hb_uv(short partid)
797 {
798 }
799
800 static void
801 xpc_disallow_hb_uv(short partid)
802 {
803 }
804
805 static void
806 xpc_disallow_all_hbs_uv(void)
807 {
808 }
809
810 static void
811 xpc_increment_heartbeat_uv(void)
812 {
813         xpc_heartbeat_uv->value++;
814 }
815
816 static void
817 xpc_offline_heartbeat_uv(void)
818 {
819         xpc_increment_heartbeat_uv();
820         xpc_heartbeat_uv->offline = 1;
821 }
822
823 static void
824 xpc_online_heartbeat_uv(void)
825 {
826         xpc_increment_heartbeat_uv();
827         xpc_heartbeat_uv->offline = 0;
828 }
829
830 static void
831 xpc_heartbeat_init_uv(void)
832 {
833         xpc_heartbeat_uv->value = 1;
834         xpc_heartbeat_uv->offline = 0;
835 }
836
837 static void
838 xpc_heartbeat_exit_uv(void)
839 {
840         xpc_offline_heartbeat_uv();
841 }
842
843 static enum xp_retval
844 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
845 {
846         struct xpc_partition_uv *part_uv = &part->sn.uv;
847         enum xp_retval ret;
848
849         ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
850                                part_uv->heartbeat_gpa,
851                                sizeof(struct xpc_heartbeat_uv));
852         if (ret != xpSuccess)
853                 return ret;
854
855         if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
856             !part_uv->cached_heartbeat.offline) {
857
858                 ret = xpNoHeartbeat;
859         } else {
860                 part->last_heartbeat = part_uv->cached_heartbeat.value;
861         }
862         return ret;
863 }
864
865 static void
866 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
867                                     unsigned long remote_rp_gpa, int nasid)
868 {
869         short partid = remote_rp->SAL_partid;
870         struct xpc_partition *part = &xpc_partitions[partid];
871         struct xpc_activate_mq_msg_activate_req_uv msg;
872
873         part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
874         part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
875         part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
876         part->sn.uv.activate_gru_mq_desc_gpa =
877             remote_rp->sn.uv.activate_gru_mq_desc_gpa;
878
879         /*
880          * ??? Is it a good idea to make this conditional on what is
881          * ??? potentially stale state information?
882          */
883         if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
884                 msg.rp_gpa = uv_gpa(xpc_rsvd_page);
885                 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
886                 msg.activate_gru_mq_desc_gpa =
887                     xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
888                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
889                                            XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
890         }
891
892         if (part->act_state == XPC_P_AS_INACTIVE)
893                 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
894 }
895
896 static void
897 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
898 {
899         xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
900 }
901
902 static void
903 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
904 {
905         struct xpc_activate_mq_msg_deactivate_req_uv msg;
906
907         /*
908          * ??? Is it a good idea to make this conditional on what is
909          * ??? potentially stale state information?
910          */
911         if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
912             part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
913
914                 msg.reason = part->reason;
915                 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
916                                          XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
917         }
918 }
919
920 static void
921 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
922 {
923         /* nothing needs to be done */
924         return;
925 }
926
927 static void
928 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
929 {
930         head->first = NULL;
931         head->last = NULL;
932         spin_lock_init(&head->lock);
933         head->n_entries = 0;
934 }
935
936 static void *
937 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
938 {
939         unsigned long irq_flags;
940         struct xpc_fifo_entry_uv *first;
941
942         spin_lock_irqsave(&head->lock, irq_flags);
943         first = head->first;
944         if (head->first != NULL) {
945                 head->first = first->next;
946                 if (head->first == NULL)
947                         head->last = NULL;
948         }
949         head->n_entries--;
950         BUG_ON(head->n_entries < 0);
951         spin_unlock_irqrestore(&head->lock, irq_flags);
952         first->next = NULL;
953         return first;
954 }
955
956 static void
957 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
958                       struct xpc_fifo_entry_uv *last)
959 {
960         unsigned long irq_flags;
961
962         last->next = NULL;
963         spin_lock_irqsave(&head->lock, irq_flags);
964         if (head->last != NULL)
965                 head->last->next = last;
966         else
967                 head->first = last;
968         head->last = last;
969         head->n_entries++;
970         spin_unlock_irqrestore(&head->lock, irq_flags);
971 }
972
973 static int
974 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
975 {
976         return head->n_entries;
977 }
978
979 /*
980  * Setup the channel structures that are uv specific.
981  */
982 static enum xp_retval
983 xpc_setup_ch_structures_uv(struct xpc_partition *part)
984 {
985         struct xpc_channel_uv *ch_uv;
986         int ch_number;
987
988         for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
989                 ch_uv = &part->channels[ch_number].sn.uv;
990
991                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
992                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
993         }
994
995         return xpSuccess;
996 }
997
998 /*
999  * Teardown the channel structures that are uv specific.
1000  */
1001 static void
1002 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1003 {
1004         /* nothing needs to be done */
1005         return;
1006 }
1007
1008 static enum xp_retval
1009 xpc_make_first_contact_uv(struct xpc_partition *part)
1010 {
1011         struct xpc_activate_mq_msg_uv msg;
1012
1013         /*
1014          * We send a sync msg to get the remote partition's remote_act_state
1015          * updated to our current act_state which at this point should
1016          * be XPC_P_AS_ACTIVATING.
1017          */
1018         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1019                                       XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1020
1021         while (part->sn.uv.remote_act_state != XPC_P_AS_ACTIVATING) {
1022
1023                 dev_dbg(xpc_part, "waiting to make first contact with "
1024                         "partition %d\n", XPC_PARTID(part));
1025
1026                 /* wait a 1/4 of a second or so */
1027                 (void)msleep_interruptible(250);
1028
1029                 if (part->act_state == XPC_P_AS_DEACTIVATING)
1030                         return part->reason;
1031         }
1032
1033         return xpSuccess;
1034 }
1035
1036 static u64
1037 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1038 {
1039         unsigned long irq_flags;
1040         union xpc_channel_ctl_flags chctl;
1041
1042         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1043         chctl = part->chctl;
1044         if (chctl.all_flags != 0)
1045                 part->chctl.all_flags = 0;
1046
1047         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1048         return chctl.all_flags;
1049 }
1050
1051 static enum xp_retval
1052 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1053 {
1054         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1055         struct xpc_send_msg_slot_uv *msg_slot;
1056         unsigned long irq_flags;
1057         int nentries;
1058         int entry;
1059         size_t nbytes;
1060
1061         for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1062                 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1063                 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1064                 if (ch_uv->send_msg_slots == NULL)
1065                         continue;
1066
1067                 for (entry = 0; entry < nentries; entry++) {
1068                         msg_slot = &ch_uv->send_msg_slots[entry];
1069
1070                         msg_slot->msg_slot_number = entry;
1071                         xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1072                                               &msg_slot->next);
1073                 }
1074
1075                 spin_lock_irqsave(&ch->lock, irq_flags);
1076                 if (nentries < ch->local_nentries)
1077                         ch->local_nentries = nentries;
1078                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1079                 return xpSuccess;
1080         }
1081
1082         return xpNoMemory;
1083 }
1084
1085 static enum xp_retval
1086 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1087 {
1088         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1089         struct xpc_notify_mq_msg_uv *msg_slot;
1090         unsigned long irq_flags;
1091         int nentries;
1092         int entry;
1093         size_t nbytes;
1094
1095         for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1096                 nbytes = nentries * ch->entry_size;
1097                 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1098                 if (ch_uv->recv_msg_slots == NULL)
1099                         continue;
1100
1101                 for (entry = 0; entry < nentries; entry++) {
1102                         msg_slot = ch_uv->recv_msg_slots +
1103                             entry * ch->entry_size;
1104
1105                         msg_slot->hdr.msg_slot_number = entry;
1106                 }
1107
1108                 spin_lock_irqsave(&ch->lock, irq_flags);
1109                 if (nentries < ch->remote_nentries)
1110                         ch->remote_nentries = nentries;
1111                 spin_unlock_irqrestore(&ch->lock, irq_flags);
1112                 return xpSuccess;
1113         }
1114
1115         return xpNoMemory;
1116 }
1117
1118 /*
1119  * Allocate msg_slots associated with the channel.
1120  */
1121 static enum xp_retval
1122 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1123 {
1124         static enum xp_retval ret;
1125         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1126
1127         DBUG_ON(ch->flags & XPC_C_SETUP);
1128
1129         ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1130                                                    gru_message_queue_desc),
1131                                                    GFP_KERNEL);
1132         if (ch_uv->cached_notify_gru_mq_desc == NULL)
1133                 return xpNoMemory;
1134
1135         ret = xpc_allocate_send_msg_slot_uv(ch);
1136         if (ret == xpSuccess) {
1137
1138                 ret = xpc_allocate_recv_msg_slot_uv(ch);
1139                 if (ret != xpSuccess) {
1140                         kfree(ch_uv->send_msg_slots);
1141                         xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1142                 }
1143         }
1144         return ret;
1145 }
1146
1147 /*
1148  * Free up msg_slots and clear other stuff that were setup for the specified
1149  * channel.
1150  */
1151 static void
1152 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1153 {
1154         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1155
1156         DBUG_ON(!spin_is_locked(&ch->lock));
1157
1158         kfree(ch_uv->cached_notify_gru_mq_desc);
1159         ch_uv->cached_notify_gru_mq_desc = NULL;
1160
1161         if (ch->flags & XPC_C_SETUP) {
1162                 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1163                 kfree(ch_uv->send_msg_slots);
1164                 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1165                 kfree(ch_uv->recv_msg_slots);
1166         }
1167 }
1168
1169 static void
1170 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1171 {
1172         struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1173
1174         msg.ch_number = ch->number;
1175         msg.reason = ch->reason;
1176         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1177                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1178 }
1179
1180 static void
1181 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1182 {
1183         struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1184
1185         msg.ch_number = ch->number;
1186         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1187                                     XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1188 }
1189
1190 static void
1191 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1192 {
1193         struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1194
1195         msg.ch_number = ch->number;
1196         msg.entry_size = ch->entry_size;
1197         msg.local_nentries = ch->local_nentries;
1198         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1199                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1200 }
1201
1202 static void
1203 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1204 {
1205         struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1206
1207         msg.ch_number = ch->number;
1208         msg.local_nentries = ch->local_nentries;
1209         msg.remote_nentries = ch->remote_nentries;
1210         msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1211         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1212                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1213 }
1214
1215 static void
1216 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1217 {
1218         struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1219
1220         msg.ch_number = ch->number;
1221         xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1222                                     XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1223 }
1224
1225 static void
1226 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1227 {
1228         unsigned long irq_flags;
1229
1230         spin_lock_irqsave(&part->chctl_lock, irq_flags);
1231         part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1232         spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1233
1234         xpc_wakeup_channel_mgr(part);
1235 }
1236
1237 static enum xp_retval
1238 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1239                                unsigned long gru_mq_desc_gpa)
1240 {
1241         struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1242
1243         DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1244         return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1245                                                gru_mq_desc_gpa);
1246 }
1247
1248 static void
1249 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1250 {
1251         struct xpc_activate_mq_msg_uv msg;
1252
1253         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1254                                       XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1255 }
1256
1257 static void
1258 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1259 {
1260         struct xpc_activate_mq_msg_uv msg;
1261
1262         xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1263                                       XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1264 }
1265
1266 static void
1267 xpc_assume_partition_disengaged_uv(short partid)
1268 {
1269         struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1270         unsigned long irq_flags;
1271
1272         spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1273         part_uv->flags &= ~XPC_P_ENGAGED_UV;
1274         spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1275 }
1276
1277 static int
1278 xpc_partition_engaged_uv(short partid)
1279 {
1280         return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1281 }
1282
1283 static int
1284 xpc_any_partition_engaged_uv(void)
1285 {
1286         struct xpc_partition_uv *part_uv;
1287         short partid;
1288
1289         for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1290                 part_uv = &xpc_partitions[partid].sn.uv;
1291                 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1292                         return 1;
1293         }
1294         return 0;
1295 }
1296
1297 static enum xp_retval
1298 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1299                          struct xpc_send_msg_slot_uv **address_of_msg_slot)
1300 {
1301         enum xp_retval ret;
1302         struct xpc_send_msg_slot_uv *msg_slot;
1303         struct xpc_fifo_entry_uv *entry;
1304
1305         while (1) {
1306                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1307                 if (entry != NULL)
1308                         break;
1309
1310                 if (flags & XPC_NOWAIT)
1311                         return xpNoWait;
1312
1313                 ret = xpc_allocate_msg_wait(ch);
1314                 if (ret != xpInterrupted && ret != xpTimeout)
1315                         return ret;
1316         }
1317
1318         msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1319         *address_of_msg_slot = msg_slot;
1320         return xpSuccess;
1321 }
1322
1323 static void
1324 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1325                      struct xpc_send_msg_slot_uv *msg_slot)
1326 {
1327         xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1328
1329         /* wakeup anyone waiting for a free msg slot */
1330         if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1331                 wake_up(&ch->msg_allocate_wq);
1332 }
1333
1334 static void
1335 xpc_notify_sender_uv(struct xpc_channel *ch,
1336                      struct xpc_send_msg_slot_uv *msg_slot,
1337                      enum xp_retval reason)
1338 {
1339         xpc_notify_func func = msg_slot->func;
1340
1341         if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1342
1343                 atomic_dec(&ch->n_to_notify);
1344
1345                 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1346                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1347                         msg_slot->msg_slot_number, ch->partid, ch->number);
1348
1349                 func(reason, ch->partid, ch->number, msg_slot->key);
1350
1351                 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1352                         "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1353                         msg_slot->msg_slot_number, ch->partid, ch->number);
1354         }
1355 }
1356
1357 static void
1358 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1359                             struct xpc_notify_mq_msg_uv *msg)
1360 {
1361         struct xpc_send_msg_slot_uv *msg_slot;
1362         int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1363
1364         msg_slot = &ch->sn.uv.send_msg_slots[entry];
1365
1366         BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1367         msg_slot->msg_slot_number += ch->local_nentries;
1368
1369         if (msg_slot->func != NULL)
1370                 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1371
1372         xpc_free_msg_slot_uv(ch, msg_slot);
1373 }
1374
1375 static void
1376 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1377                             struct xpc_notify_mq_msg_uv *msg)
1378 {
1379         struct xpc_partition_uv *part_uv = &part->sn.uv;
1380         struct xpc_channel *ch;
1381         struct xpc_channel_uv *ch_uv;
1382         struct xpc_notify_mq_msg_uv *msg_slot;
1383         unsigned long irq_flags;
1384         int ch_number = msg->hdr.ch_number;
1385
1386         if (unlikely(ch_number >= part->nchannels)) {
1387                 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1388                         "channel number=0x%x in message from partid=%d\n",
1389                         ch_number, XPC_PARTID(part));
1390
1391                 /* get hb checker to deactivate from the remote partition */
1392                 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1393                 if (part_uv->act_state_req == 0)
1394                         xpc_activate_IRQ_rcvd++;
1395                 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1396                 part_uv->reason = xpBadChannelNumber;
1397                 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1398
1399                 wake_up_interruptible(&xpc_activate_IRQ_wq);
1400                 return;
1401         }
1402
1403         ch = &part->channels[ch_number];
1404         xpc_msgqueue_ref(ch);
1405
1406         if (!(ch->flags & XPC_C_CONNECTED)) {
1407                 xpc_msgqueue_deref(ch);
1408                 return;
1409         }
1410
1411         /* see if we're really dealing with an ACK for a previously sent msg */
1412         if (msg->hdr.size == 0) {
1413                 xpc_handle_notify_mq_ack_uv(ch, msg);
1414                 xpc_msgqueue_deref(ch);
1415                 return;
1416         }
1417
1418         /* we're dealing with a normal message sent via the notify_mq */
1419         ch_uv = &ch->sn.uv;
1420
1421         msg_slot = ch_uv->recv_msg_slots +
1422             (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1423
1424         BUG_ON(msg->hdr.msg_slot_number != msg_slot->hdr.msg_slot_number);
1425         BUG_ON(msg_slot->hdr.size != 0);
1426
1427         memcpy(msg_slot, msg, msg->hdr.size);
1428
1429         xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1430
1431         if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1432                 /*
1433                  * If there is an existing idle kthread get it to deliver
1434                  * the payload, otherwise we'll have to get the channel mgr
1435                  * for this partition to create a kthread to do the delivery.
1436                  */
1437                 if (atomic_read(&ch->kthreads_idle) > 0)
1438                         wake_up_nr(&ch->idle_wq, 1);
1439                 else
1440                         xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1441         }
1442         xpc_msgqueue_deref(ch);
1443 }
1444
1445 static irqreturn_t
1446 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1447 {
1448         struct xpc_notify_mq_msg_uv *msg;
1449         short partid;
1450         struct xpc_partition *part;
1451
1452         while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1453                NULL) {
1454
1455                 partid = msg->hdr.partid;
1456                 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1457                         dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1458                                 "invalid partid=0x%x in message\n", partid);
1459                 } else {
1460                         part = &xpc_partitions[partid];
1461
1462                         if (xpc_part_ref(part)) {
1463                                 xpc_handle_notify_mq_msg_uv(part, msg);
1464                                 xpc_part_deref(part);
1465                         }
1466                 }
1467
1468                 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1469         }
1470
1471         return IRQ_HANDLED;
1472 }
1473
1474 static int
1475 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1476 {
1477         return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1478 }
1479
1480 static void
1481 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1482 {
1483         struct xpc_channel *ch = &part->channels[ch_number];
1484         int ndeliverable_payloads;
1485
1486         xpc_msgqueue_ref(ch);
1487
1488         ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1489
1490         if (ndeliverable_payloads > 0 &&
1491             (ch->flags & XPC_C_CONNECTED) &&
1492             (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1493
1494                 xpc_activate_kthreads(ch, ndeliverable_payloads);
1495         }
1496
1497         xpc_msgqueue_deref(ch);
1498 }
1499
1500 static enum xp_retval
1501 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1502                     u16 payload_size, u8 notify_type, xpc_notify_func func,
1503                     void *key)
1504 {
1505         enum xp_retval ret = xpSuccess;
1506         struct xpc_send_msg_slot_uv *msg_slot = NULL;
1507         struct xpc_notify_mq_msg_uv *msg;
1508         u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1509         size_t msg_size;
1510
1511         DBUG_ON(notify_type != XPC_N_CALL);
1512
1513         msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1514         if (msg_size > ch->entry_size)
1515                 return xpPayloadTooBig;
1516
1517         xpc_msgqueue_ref(ch);
1518
1519         if (ch->flags & XPC_C_DISCONNECTING) {
1520                 ret = ch->reason;
1521                 goto out_1;
1522         }
1523         if (!(ch->flags & XPC_C_CONNECTED)) {
1524                 ret = xpNotConnected;
1525                 goto out_1;
1526         }
1527
1528         ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1529         if (ret != xpSuccess)
1530                 goto out_1;
1531
1532         if (func != NULL) {
1533                 atomic_inc(&ch->n_to_notify);
1534
1535                 msg_slot->key = key;
1536                 smp_wmb(); /* a non-NULL func must hit memory after the key */
1537                 msg_slot->func = func;
1538
1539                 if (ch->flags & XPC_C_DISCONNECTING) {
1540                         ret = ch->reason;
1541                         goto out_2;
1542                 }
1543         }
1544
1545         msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1546         msg->hdr.partid = xp_partition_id;
1547         msg->hdr.ch_number = ch->number;
1548         msg->hdr.size = msg_size;
1549         msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1550         memcpy(&msg->payload, payload, payload_size);
1551
1552         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1553                                msg_size);
1554         if (ret == xpSuccess)
1555                 goto out_1;
1556
1557         XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1558 out_2:
1559         if (func != NULL) {
1560                 /*
1561                  * Try to NULL the msg_slot's func field. If we fail, then
1562                  * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1563                  * case we need to pretend we succeeded to send the message
1564                  * since the user will get a callout for the disconnect error
1565                  * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1566                  * error returned here will confuse them. Additionally, since
1567                  * in this case the channel is being disconnected we don't need
1568                  * to put the the msg_slot back on the free list.
1569                  */
1570                 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1571                         ret = xpSuccess;
1572                         goto out_1;
1573                 }
1574
1575                 msg_slot->key = NULL;
1576                 atomic_dec(&ch->n_to_notify);
1577         }
1578         xpc_free_msg_slot_uv(ch, msg_slot);
1579 out_1:
1580         xpc_msgqueue_deref(ch);
1581         return ret;
1582 }
1583
1584 /*
1585  * Tell the callers of xpc_send_notify() that the status of their payloads
1586  * is unknown because the channel is now disconnecting.
1587  *
1588  * We don't worry about putting these msg_slots on the free list since the
1589  * msg_slots themselves are about to be kfree'd.
1590  */
1591 static void
1592 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1593 {
1594         struct xpc_send_msg_slot_uv *msg_slot;
1595         int entry;
1596
1597         DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1598
1599         for (entry = 0; entry < ch->local_nentries; entry++) {
1600
1601                 if (atomic_read(&ch->n_to_notify) == 0)
1602                         break;
1603
1604                 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1605                 if (msg_slot->func != NULL)
1606                         xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1607         }
1608 }
1609
1610 /*
1611  * Get the next deliverable message's payload.
1612  */
1613 static void *
1614 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1615 {
1616         struct xpc_fifo_entry_uv *entry;
1617         struct xpc_notify_mq_msg_uv *msg;
1618         void *payload = NULL;
1619
1620         if (!(ch->flags & XPC_C_DISCONNECTING)) {
1621                 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1622                 if (entry != NULL) {
1623                         msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1624                                            hdr.u.next);
1625                         payload = &msg->payload;
1626                 }
1627         }
1628         return payload;
1629 }
1630
1631 static void
1632 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1633 {
1634         struct xpc_notify_mq_msg_uv *msg;
1635         enum xp_retval ret;
1636
1637         msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1638
1639         /* return an ACK to the sender of this message */
1640
1641         msg->hdr.partid = xp_partition_id;
1642         msg->hdr.size = 0;      /* size of zero indicates this is an ACK */
1643
1644         ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1645                                sizeof(struct xpc_notify_mq_msghdr_uv));
1646         if (ret != xpSuccess)
1647                 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1648
1649         msg->hdr.msg_slot_number += ch->remote_nentries;
1650 }
1651
1652 static struct xpc_arch_operations xpc_arch_ops_uv = {
1653         .setup_partitions = xpc_setup_partitions_uv,
1654         .teardown_partitions = xpc_teardown_partitions_uv,
1655         .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1656         .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1657         .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1658
1659         .allow_hb = xpc_allow_hb_uv,
1660         .disallow_hb = xpc_disallow_hb_uv,
1661         .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1662         .increment_heartbeat = xpc_increment_heartbeat_uv,
1663         .offline_heartbeat = xpc_offline_heartbeat_uv,
1664         .online_heartbeat = xpc_online_heartbeat_uv,
1665         .heartbeat_init = xpc_heartbeat_init_uv,
1666         .heartbeat_exit = xpc_heartbeat_exit_uv,
1667         .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1668
1669         .request_partition_activation =
1670                 xpc_request_partition_activation_uv,
1671         .request_partition_reactivation =
1672                 xpc_request_partition_reactivation_uv,
1673         .request_partition_deactivation =
1674                 xpc_request_partition_deactivation_uv,
1675         .cancel_partition_deactivation_request =
1676                 xpc_cancel_partition_deactivation_request_uv,
1677
1678         .setup_ch_structures = xpc_setup_ch_structures_uv,
1679         .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1680
1681         .make_first_contact = xpc_make_first_contact_uv,
1682
1683         .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1684         .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1685         .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1686         .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1687         .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1688         .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1689         .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1690
1691         .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1692
1693         .setup_msg_structures = xpc_setup_msg_structures_uv,
1694         .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1695
1696         .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1697         .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1698         .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1699         .partition_engaged = xpc_partition_engaged_uv,
1700         .any_partition_engaged = xpc_any_partition_engaged_uv,
1701
1702         .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1703         .send_payload = xpc_send_payload_uv,
1704         .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1705         .received_payload = xpc_received_payload_uv,
1706         .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1707 };
1708
1709 int
1710 xpc_init_uv(void)
1711 {
1712         xpc_arch_ops = xpc_arch_ops_uv;
1713
1714         if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1715                 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1716                         XPC_MSG_HDR_MAX_SIZE);
1717                 return -E2BIG;
1718         }
1719
1720         xpc_activate_mq_uv = xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, 0,
1721                                                   XPC_ACTIVATE_IRQ_NAME,
1722                                                   xpc_handle_activate_IRQ_uv);
1723         if (IS_ERR(xpc_activate_mq_uv))
1724                 return PTR_ERR(xpc_activate_mq_uv);
1725
1726         xpc_notify_mq_uv = xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, 0,
1727                                                 XPC_NOTIFY_IRQ_NAME,
1728                                                 xpc_handle_notify_IRQ_uv);
1729         if (IS_ERR(xpc_notify_mq_uv)) {
1730                 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1731                 return PTR_ERR(xpc_notify_mq_uv);
1732         }
1733
1734         return 0;
1735 }
1736
1737 void
1738 xpc_exit_uv(void)
1739 {
1740         xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1741         xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1742 }