]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - drivers/misc/sgi-xp/xpc_uv.c
projects / mirror_ubuntu-jammy-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
mm: replace all open encodings for NUMA_NO_NODE
[mirror_ubuntu-jammy-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/cpu.h>
22 #include <linux/module.h>
23 #include <linux/err.h>
24 #include <linux/slab.h>
25 #include <linux/numa.h>
26 #include <asm/uv/uv_hub.h>
27 #if defined CONFIG_X86_64
28 #include <asm/uv/bios.h>
29 #include <asm/uv/uv_irq.h>
30 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
31 #include <asm/sn/intr.h>
32 #include <asm/sn/sn_sal.h>
33 #endif
34 #include "../sgi-gru/gru.h"
35 #include "../sgi-gru/grukservices.h"
36 #include "xpc.h"
37
38 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
39 struct uv_IO_APIC_route_entry {
40 __u64 vector : 8,
41 delivery_mode : 3,
42 dest_mode : 1,
43 delivery_status : 1,
44 polarity : 1,
45 __reserved_1 : 1,
46 trigger : 1,
47 mask : 1,
48 __reserved_2 : 15,
49 dest : 32;
50 };
51 #endif
52
53 static struct xpc_heartbeat_uv *xpc_heartbeat_uv;
54
55 #define XPC_ACTIVATE_MSG_SIZE_UV (1 * GRU_CACHE_LINE_BYTES)
56 #define XPC_ACTIVATE_MQ_SIZE_UV (4 * XP_MAX_NPARTITIONS_UV * \
57 XPC_ACTIVATE_MSG_SIZE_UV)
58 #define XPC_ACTIVATE_IRQ_NAME "xpc_activate"
59
60 #define XPC_NOTIFY_MSG_SIZE_UV (2 * GRU_CACHE_LINE_BYTES)
61 #define XPC_NOTIFY_MQ_SIZE_UV (4 * XP_MAX_NPARTITIONS_UV * \
62 XPC_NOTIFY_MSG_SIZE_UV)
63 #define XPC_NOTIFY_IRQ_NAME "xpc_notify"
64
65 static int xpc_mq_node = NUMA_NO_NODE;
66
67 static struct xpc_gru_mq_uv *xpc_activate_mq_uv;
68 static struct xpc_gru_mq_uv *xpc_notify_mq_uv;
69
70 static int
71 xpc_setup_partitions_uv(void)
72 {
73 short partid;
74 struct xpc_partition_uv *part_uv;
75
76 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
77 part_uv = &xpc_partitions[partid].sn.uv;
78
79 mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
80 spin_lock_init(&part_uv->flags_lock);
81 part_uv->remote_act_state = XPC_P_AS_INACTIVE;
82 }
83 return 0;
84 }
85
86 static void
87 xpc_teardown_partitions_uv(void)
88 {
89 short partid;
90 struct xpc_partition_uv *part_uv;
91 unsigned long irq_flags;
92
93 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
94 part_uv = &xpc_partitions[partid].sn.uv;
95
96 if (part_uv->cached_activate_gru_mq_desc != NULL) {
97 mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
98 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
99 part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
100 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
101 kfree(part_uv->cached_activate_gru_mq_desc);
102 part_uv->cached_activate_gru_mq_desc = NULL;
103 mutex_unlock(&part_uv->
104 cached_activate_gru_mq_desc_mutex);
105 }
106 }
107 }
108
109 static int
110 xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
111 {
112 int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
113
114 #if defined CONFIG_X86_64
115 mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset,
116 UV_AFFINITY_CPU);
117 if (mq->irq < 0)
118 return mq->irq;
119
120 mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
121
122 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
123 if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
124 mq->irq = SGI_XPC_ACTIVATE;
125 else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
126 mq->irq = SGI_XPC_NOTIFY;
127 else
128 return -EINVAL;
129
130 mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
131 uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
132 #else
133 #error not a supported configuration
134 #endif
135
136 return 0;
137 }
138
139 static void
140 xpc_release_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq)
141 {
142 #if defined CONFIG_X86_64
143 uv_teardown_irq(mq->irq);
144
145 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
146 int mmr_pnode;
147 unsigned long mmr_value;
148
149 mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
150 mmr_value = 1UL << 16;
151
152 uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
153 #else
154 #error not a supported configuration
155 #endif
156 }
157
158 static int
159 xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
160 {
161 int ret;
162
163 #if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
164 int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
165
166 ret = sn_mq_watchlist_alloc(mmr_pnode, (void *)uv_gpa(mq->address),
167 mq->order, &mq->mmr_offset);
168 if (ret < 0) {
169 dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
170 ret);
171 return -EBUSY;
172 }
173 #elif defined CONFIG_X86_64
174 ret = uv_bios_mq_watchlist_alloc(uv_gpa(mq->address),
175 mq->order, &mq->mmr_offset);
176 if (ret < 0) {
177 dev_err(xpc_part, "uv_bios_mq_watchlist_alloc() failed, "
178 "ret=%d\n", ret);
179 return ret;
180 }
181 #else
182 #error not a supported configuration
183 #endif
184
185 mq->watchlist_num = ret;
186 return 0;
187 }
188
189 static void
190 xpc_gru_mq_watchlist_free_uv(struct xpc_gru_mq_uv *mq)
191 {
192 int ret;
193 int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
194
195 #if defined CONFIG_X86_64
196 ret = uv_bios_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
197 BUG_ON(ret != BIOS_STATUS_SUCCESS);
198 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
199 ret = sn_mq_watchlist_free(mmr_pnode, mq->watchlist_num);
200 BUG_ON(ret != SALRET_OK);
201 #else
202 #error not a supported configuration
203 #endif
204 }
205
206 static struct xpc_gru_mq_uv *
207 xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
208 irq_handler_t irq_handler)
209 {
210 enum xp_retval xp_ret;
211 int ret;
212 int nid;
213 int nasid;
214 int pg_order;
215 struct page *page;
216 struct xpc_gru_mq_uv *mq;
217 struct uv_IO_APIC_route_entry *mmr_value;
218
219 mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
220 if (mq == NULL) {
221 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
222 "a xpc_gru_mq_uv structure\n");
223 ret = -ENOMEM;
224 goto out_0;
225 }
226
227 mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
228 GFP_KERNEL);
229 if (mq->gru_mq_desc == NULL) {
230 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
231 "a gru_message_queue_desc structure\n");
232 ret = -ENOMEM;
233 goto out_1;
234 }
235
236 pg_order = get_order(mq_size);
237 mq->order = pg_order + PAGE_SHIFT;
238 mq_size = 1UL << mq->order;
239
240 mq->mmr_blade = uv_cpu_to_blade_id(cpu);
241
242 nid = cpu_to_node(cpu);
243 page = __alloc_pages_node(nid,
244 GFP_KERNEL | __GFP_ZERO | __GFP_THISNODE,
245 pg_order);
246 if (page == NULL) {
247 dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to alloc %d "
248 "bytes of memory on nid=%d for GRU mq\n", mq_size, nid);
249 ret = -ENOMEM;
250 goto out_2;
251 }
252 mq->address = page_address(page);
253
254 /* enable generation of irq when GRU mq operation occurs to this mq */
255 ret = xpc_gru_mq_watchlist_alloc_uv(mq);
256 if (ret != 0)
257 goto out_3;
258
259 ret = xpc_get_gru_mq_irq_uv(mq, cpu, irq_name);
260 if (ret != 0)
261 goto out_4;
262
263 ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
264 if (ret != 0) {
265 dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
266 mq->irq, -ret);
267 goto out_5;
268 }
269
270 nasid = UV_PNODE_TO_NASID(uv_cpu_to_pnode(cpu));
271
272 mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
273 ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
274 nasid, mmr_value->vector, mmr_value->dest);
275 if (ret != 0) {
276 dev_err(xpc_part, "gru_create_message_queue() returned "
277 "error=%d\n", ret);
278 ret = -EINVAL;
279 goto out_6;
280 }
281
282 /* allow other partitions to access this GRU mq */
283 xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
284 if (xp_ret != xpSuccess) {
285 ret = -EACCES;
286 goto out_6;
287 }
288
289 return mq;
290
291 /* something went wrong */
292 out_6:
293 free_irq(mq->irq, NULL);
294 out_5:
295 xpc_release_gru_mq_irq_uv(mq);
296 out_4:
297 xpc_gru_mq_watchlist_free_uv(mq);
298 out_3:
299 free_pages((unsigned long)mq->address, pg_order);
300 out_2:
301 kfree(mq->gru_mq_desc);
302 out_1:
303 kfree(mq);
304 out_0:
305 return ERR_PTR(ret);
306 }
307
308 static void
309 xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
310 {
311 unsigned int mq_size;
312 int pg_order;
313 int ret;
314
315 /* disallow other partitions to access GRU mq */
316 mq_size = 1UL << mq->order;
317 ret = xp_restrict_memprotect(xp_pa(mq->address), mq_size);
318 BUG_ON(ret != xpSuccess);
319
320 /* unregister irq handler and release mq irq/vector mapping */
321 free_irq(mq->irq, NULL);
322 xpc_release_gru_mq_irq_uv(mq);
323
324 /* disable generation of irq when GRU mq op occurs to this mq */
325 xpc_gru_mq_watchlist_free_uv(mq);
326
327 pg_order = mq->order - PAGE_SHIFT;
328 free_pages((unsigned long)mq->address, pg_order);
329
330 kfree(mq);
331 }
332
333 static enum xp_retval
334 xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
335 size_t msg_size)
336 {
337 enum xp_retval xp_ret;
338 int ret;
339
340 while (1) {
341 ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
342 if (ret == MQE_OK) {
343 xp_ret = xpSuccess;
344 break;
345 }
346
347 if (ret == MQE_QUEUE_FULL) {
348 dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
349 "error=MQE_QUEUE_FULL\n");
350 /* !!! handle QLimit reached; delay & try again */
351 /* ??? Do we add a limit to the number of retries? */
352 (void)msleep_interruptible(10);
353 } else if (ret == MQE_CONGESTION) {
354 dev_dbg(xpc_chan, "gru_send_message_gpa() returned "
355 "error=MQE_CONGESTION\n");
356 /* !!! handle LB Overflow; simply try again */
357 /* ??? Do we add a limit to the number of retries? */
358 } else {
359 /* !!! Currently this is MQE_UNEXPECTED_CB_ERR */
360 dev_err(xpc_chan, "gru_send_message_gpa() returned "
361 "error=%d\n", ret);
362 xp_ret = xpGruSendMqError;
363 break;
364 }
365 }
366 return xp_ret;
367 }
368
369 static void
370 xpc_process_activate_IRQ_rcvd_uv(void)
371 {
372 unsigned long irq_flags;
373 short partid;
374 struct xpc_partition *part;
375 u8 act_state_req;
376
377 DBUG_ON(xpc_activate_IRQ_rcvd == 0);
378
379 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
380 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
381 part = &xpc_partitions[partid];
382
383 if (part->sn.uv.act_state_req == 0)
384 continue;
385
386 xpc_activate_IRQ_rcvd--;
387 BUG_ON(xpc_activate_IRQ_rcvd < 0);
388
389 act_state_req = part->sn.uv.act_state_req;
390 part->sn.uv.act_state_req = 0;
391 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
392
393 if (act_state_req == XPC_P_ASR_ACTIVATE_UV) {
394 if (part->act_state == XPC_P_AS_INACTIVE)
395 xpc_activate_partition(part);
396 else if (part->act_state == XPC_P_AS_DEACTIVATING)
397 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
398
399 } else if (act_state_req == XPC_P_ASR_REACTIVATE_UV) {
400 if (part->act_state == XPC_P_AS_INACTIVE)
401 xpc_activate_partition(part);
402 else
403 XPC_DEACTIVATE_PARTITION(part, xpReactivating);
404
405 } else if (act_state_req == XPC_P_ASR_DEACTIVATE_UV) {
406 XPC_DEACTIVATE_PARTITION(part, part->sn.uv.reason);
407
408 } else {
409 BUG();
410 }
411
412 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
413 if (xpc_activate_IRQ_rcvd == 0)
414 break;
415 }
416 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
417
418 }
419
420 static void
421 xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
422 struct xpc_activate_mq_msghdr_uv *msg_hdr,
423 int part_setup,
424 int *wakeup_hb_checker)
425 {
426 unsigned long irq_flags;
427 struct xpc_partition_uv *part_uv = &part->sn.uv;
428 struct xpc_openclose_args *args;
429
430 part_uv->remote_act_state = msg_hdr->act_state;
431
432 switch (msg_hdr->type) {
433 case XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV:
434 /* syncing of remote_act_state was just done above */
435 break;
436
437 case XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV: {
438 struct xpc_activate_mq_msg_activate_req_uv *msg;
439
440 /*
441 * ??? Do we deal here with ts_jiffies being different
442 * ??? if act_state != XPC_P_AS_INACTIVE instead of
443 * ??? below?
444 */
445 msg = container_of(msg_hdr, struct
446 xpc_activate_mq_msg_activate_req_uv, hdr);
447
448 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
449 if (part_uv->act_state_req == 0)
450 xpc_activate_IRQ_rcvd++;
451 part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
452 part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
453 part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
454 part_uv->heartbeat_gpa = msg->heartbeat_gpa;
455
456 if (msg->activate_gru_mq_desc_gpa !=
457 part_uv->activate_gru_mq_desc_gpa) {
458 spin_lock(&part_uv->flags_lock);
459 part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
460 spin_unlock(&part_uv->flags_lock);
461 part_uv->activate_gru_mq_desc_gpa =
462 msg->activate_gru_mq_desc_gpa;
463 }
464 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
465
466 (*wakeup_hb_checker)++;
467 break;
468 }
469 case XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV: {
470 struct xpc_activate_mq_msg_deactivate_req_uv *msg;
471
472 msg = container_of(msg_hdr, struct
473 xpc_activate_mq_msg_deactivate_req_uv, hdr);
474
475 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
476 if (part_uv->act_state_req == 0)
477 xpc_activate_IRQ_rcvd++;
478 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
479 part_uv->reason = msg->reason;
480 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
481
482 (*wakeup_hb_checker)++;
483 return;
484 }
485 case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV: {
486 struct xpc_activate_mq_msg_chctl_closerequest_uv *msg;
487
488 if (!part_setup)
489 break;
490
491 msg = container_of(msg_hdr, struct
492 xpc_activate_mq_msg_chctl_closerequest_uv,
493 hdr);
494 args = &part->remote_openclose_args[msg->ch_number];
495 args->reason = msg->reason;
496
497 spin_lock_irqsave(&part->chctl_lock, irq_flags);
498 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREQUEST;
499 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
500
501 xpc_wakeup_channel_mgr(part);
502 break;
503 }
504 case XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV: {
505 struct xpc_activate_mq_msg_chctl_closereply_uv *msg;
506
507 if (!part_setup)
508 break;
509
510 msg = container_of(msg_hdr, struct
511 xpc_activate_mq_msg_chctl_closereply_uv,
512 hdr);
513
514 spin_lock_irqsave(&part->chctl_lock, irq_flags);
515 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_CLOSEREPLY;
516 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
517
518 xpc_wakeup_channel_mgr(part);
519 break;
520 }
521 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV: {
522 struct xpc_activate_mq_msg_chctl_openrequest_uv *msg;
523
524 if (!part_setup)
525 break;
526
527 msg = container_of(msg_hdr, struct
528 xpc_activate_mq_msg_chctl_openrequest_uv,
529 hdr);
530 args = &part->remote_openclose_args[msg->ch_number];
531 args->entry_size = msg->entry_size;
532 args->local_nentries = msg->local_nentries;
533
534 spin_lock_irqsave(&part->chctl_lock, irq_flags);
535 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREQUEST;
536 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
537
538 xpc_wakeup_channel_mgr(part);
539 break;
540 }
541 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV: {
542 struct xpc_activate_mq_msg_chctl_openreply_uv *msg;
543
544 if (!part_setup)
545 break;
546
547 msg = container_of(msg_hdr, struct
548 xpc_activate_mq_msg_chctl_openreply_uv, hdr);
549 args = &part->remote_openclose_args[msg->ch_number];
550 args->remote_nentries = msg->remote_nentries;
551 args->local_nentries = msg->local_nentries;
552 args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
553
554 spin_lock_irqsave(&part->chctl_lock, irq_flags);
555 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
556 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
557
558 xpc_wakeup_channel_mgr(part);
559 break;
560 }
561 case XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV: {
562 struct xpc_activate_mq_msg_chctl_opencomplete_uv *msg;
563
564 if (!part_setup)
565 break;
566
567 msg = container_of(msg_hdr, struct
568 xpc_activate_mq_msg_chctl_opencomplete_uv, hdr);
569 spin_lock_irqsave(&part->chctl_lock, irq_flags);
570 part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENCOMPLETE;
571 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
572
573 xpc_wakeup_channel_mgr(part);
574 }
575 case XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV:
576 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
577 part_uv->flags |= XPC_P_ENGAGED_UV;
578 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
579 break;
580
581 case XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV:
582 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
583 part_uv->flags &= ~XPC_P_ENGAGED_UV;
584 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
585 break;
586
587 default:
588 dev_err(xpc_part, "received unknown activate_mq msg type=%d "
589 "from partition=%d\n", msg_hdr->type, XPC_PARTID(part));
590
591 /* get hb checker to deactivate from the remote partition */
592 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
593 if (part_uv->act_state_req == 0)
594 xpc_activate_IRQ_rcvd++;
595 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
596 part_uv->reason = xpBadMsgType;
597 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
598
599 (*wakeup_hb_checker)++;
600 return;
601 }
602
603 if (msg_hdr->rp_ts_jiffies != part->remote_rp_ts_jiffies &&
604 part->remote_rp_ts_jiffies != 0) {
605 /*
606 * ??? Does what we do here need to be sensitive to
607 * ??? act_state or remote_act_state?
608 */
609 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
610 if (part_uv->act_state_req == 0)
611 xpc_activate_IRQ_rcvd++;
612 part_uv->act_state_req = XPC_P_ASR_REACTIVATE_UV;
613 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
614
615 (*wakeup_hb_checker)++;
616 }
617 }
618
619 static irqreturn_t
620 xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
621 {
622 struct xpc_activate_mq_msghdr_uv *msg_hdr;
623 short partid;
624 struct xpc_partition *part;
625 int wakeup_hb_checker = 0;
626 int part_referenced;
627
628 while (1) {
629 msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
630 if (msg_hdr == NULL)
631 break;
632
633 partid = msg_hdr->partid;
634 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
635 dev_err(xpc_part, "xpc_handle_activate_IRQ_uv() "
636 "received invalid partid=0x%x in message\n",
637 partid);
638 } else {
639 part = &xpc_partitions[partid];
640
641 part_referenced = xpc_part_ref(part);
642 xpc_handle_activate_mq_msg_uv(part, msg_hdr,
643 part_referenced,
644 &wakeup_hb_checker);
645 if (part_referenced)
646 xpc_part_deref(part);
647 }
648
649 gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
650 }
651
652 if (wakeup_hb_checker)
653 wake_up_interruptible(&xpc_activate_IRQ_wq);
654
655 return IRQ_HANDLED;
656 }
657
658 static enum xp_retval
659 xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
660 unsigned long gru_mq_desc_gpa)
661 {
662 enum xp_retval ret;
663
664 ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
665 sizeof(struct gru_message_queue_desc));
666 if (ret == xpSuccess)
667 gru_mq_desc->mq = NULL;
668
669 return ret;
670 }
671
672 static enum xp_retval
673 xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
674 int msg_type)
675 {
676 struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
677 struct xpc_partition_uv *part_uv = &part->sn.uv;
678 struct gru_message_queue_desc *gru_mq_desc;
679 unsigned long irq_flags;
680 enum xp_retval ret;
681
682 DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
683
684 msg_hdr->type = msg_type;
685 msg_hdr->partid = xp_partition_id;
686 msg_hdr->act_state = part->act_state;
687 msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
688
689 mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
690 again:
691 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
692 gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
693 if (gru_mq_desc == NULL) {
694 gru_mq_desc = kmalloc(sizeof(struct
695 gru_message_queue_desc),
696 GFP_KERNEL);
697 if (gru_mq_desc == NULL) {
698 ret = xpNoMemory;
699 goto done;
700 }
701 part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
702 }
703
704 ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
705 part_uv->
706 activate_gru_mq_desc_gpa);
707 if (ret != xpSuccess)
708 goto done;
709
710 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
711 part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
712 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
713 }
714
715 /* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
716 ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
717 msg_size);
718 if (ret != xpSuccess) {
719 smp_rmb(); /* ensure a fresh copy of part_uv->flags */
720 if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
721 goto again;
722 }
723 done:
724 mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
725 return ret;
726 }
727
728 static void
729 xpc_send_activate_IRQ_part_uv(struct xpc_partition *part, void *msg,
730 size_t msg_size, int msg_type)
731 {
732 enum xp_retval ret;
733
734 ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
735 if (unlikely(ret != xpSuccess))
736 XPC_DEACTIVATE_PARTITION(part, ret);
737 }
738
739 static void
740 xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
741 void *msg, size_t msg_size, int msg_type)
742 {
743 struct xpc_partition *part = &xpc_partitions[ch->partid];
744 enum xp_retval ret;
745
746 ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
747 if (unlikely(ret != xpSuccess)) {
748 if (irq_flags != NULL)
749 spin_unlock_irqrestore(&ch->lock, *irq_flags);
750
751 XPC_DEACTIVATE_PARTITION(part, ret);
752
753 if (irq_flags != NULL)
754 spin_lock_irqsave(&ch->lock, *irq_flags);
755 }
756 }
757
758 static void
759 xpc_send_local_activate_IRQ_uv(struct xpc_partition *part, int act_state_req)
760 {
761 unsigned long irq_flags;
762 struct xpc_partition_uv *part_uv = &part->sn.uv;
763
764 /*
765 * !!! Make our side think that the remote partition sent an activate
766 * !!! mq message our way by doing what the activate IRQ handler would
767 * !!! do had one really been sent.
768 */
769
770 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
771 if (part_uv->act_state_req == 0)
772 xpc_activate_IRQ_rcvd++;
773 part_uv->act_state_req = act_state_req;
774 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
775
776 wake_up_interruptible(&xpc_activate_IRQ_wq);
777 }
778
779 static enum xp_retval
780 xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
781 size_t *len)
782 {
783 s64 status;
784 enum xp_retval ret;
785
786 #if defined CONFIG_X86_64
787 status = uv_bios_reserved_page_pa((u64)buf, cookie, (u64 *)rp_pa,
788 (u64 *)len);
789 if (status == BIOS_STATUS_SUCCESS)
790 ret = xpSuccess;
791 else if (status == BIOS_STATUS_MORE_PASSES)
792 ret = xpNeedMoreInfo;
793 else
794 ret = xpBiosError;
795
796 #elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
797 status = sn_partition_reserved_page_pa((u64)buf, cookie, rp_pa, len);
798 if (status == SALRET_OK)
799 ret = xpSuccess;
800 else if (status == SALRET_MORE_PASSES)
801 ret = xpNeedMoreInfo;
802 else
803 ret = xpSalError;
804
805 #else
806 #error not a supported configuration
807 #endif
808
809 return ret;
810 }
811
812 static int
813 xpc_setup_rsvd_page_uv(struct xpc_rsvd_page *rp)
814 {
815 xpc_heartbeat_uv =
816 &xpc_partitions[sn_partition_id].sn.uv.cached_heartbeat;
817 rp->sn.uv.heartbeat_gpa = uv_gpa(xpc_heartbeat_uv);
818 rp->sn.uv.activate_gru_mq_desc_gpa =
819 uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
820 return 0;
821 }
822
823 static void
824 xpc_allow_hb_uv(short partid)
825 {
826 }
827
828 static void
829 xpc_disallow_hb_uv(short partid)
830 {
831 }
832
833 static void
834 xpc_disallow_all_hbs_uv(void)
835 {
836 }
837
838 static void
839 xpc_increment_heartbeat_uv(void)
840 {
841 xpc_heartbeat_uv->value++;
842 }
843
844 static void
845 xpc_offline_heartbeat_uv(void)
846 {
847 xpc_increment_heartbeat_uv();
848 xpc_heartbeat_uv->offline = 1;
849 }
850
851 static void
852 xpc_online_heartbeat_uv(void)
853 {
854 xpc_increment_heartbeat_uv();
855 xpc_heartbeat_uv->offline = 0;
856 }
857
858 static void
859 xpc_heartbeat_init_uv(void)
860 {
861 xpc_heartbeat_uv->value = 1;
862 xpc_heartbeat_uv->offline = 0;
863 }
864
865 static void
866 xpc_heartbeat_exit_uv(void)
867 {
868 xpc_offline_heartbeat_uv();
869 }
870
871 static enum xp_retval
872 xpc_get_remote_heartbeat_uv(struct xpc_partition *part)
873 {
874 struct xpc_partition_uv *part_uv = &part->sn.uv;
875 enum xp_retval ret;
876
877 ret = xp_remote_memcpy(uv_gpa(&part_uv->cached_heartbeat),
878 part_uv->heartbeat_gpa,
879 sizeof(struct xpc_heartbeat_uv));
880 if (ret != xpSuccess)
881 return ret;
882
883 if (part_uv->cached_heartbeat.value == part->last_heartbeat &&
884 !part_uv->cached_heartbeat.offline) {
885
886 ret = xpNoHeartbeat;
887 } else {
888 part->last_heartbeat = part_uv->cached_heartbeat.value;
889 }
890 return ret;
891 }
892
893 static void
894 xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
895 unsigned long remote_rp_gpa, int nasid)
896 {
897 short partid = remote_rp->SAL_partid;
898 struct xpc_partition *part = &xpc_partitions[partid];
899 struct xpc_activate_mq_msg_activate_req_uv msg;
900
901 part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
902 part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
903 part->sn.uv.heartbeat_gpa = remote_rp->sn.uv.heartbeat_gpa;
904 part->sn.uv.activate_gru_mq_desc_gpa =
905 remote_rp->sn.uv.activate_gru_mq_desc_gpa;
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_INACTIVE) {
912 msg.rp_gpa = uv_gpa(xpc_rsvd_page);
913 msg.heartbeat_gpa = xpc_rsvd_page->sn.uv.heartbeat_gpa;
914 msg.activate_gru_mq_desc_gpa =
915 xpc_rsvd_page->sn.uv.activate_gru_mq_desc_gpa;
916 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
917 XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
918 }
919
920 if (part->act_state == XPC_P_AS_INACTIVE)
921 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
922 }
923
924 static void
925 xpc_request_partition_reactivation_uv(struct xpc_partition *part)
926 {
927 xpc_send_local_activate_IRQ_uv(part, XPC_P_ASR_ACTIVATE_UV);
928 }
929
930 static void
931 xpc_request_partition_deactivation_uv(struct xpc_partition *part)
932 {
933 struct xpc_activate_mq_msg_deactivate_req_uv msg;
934
935 /*
936 * ??? Is it a good idea to make this conditional on what is
937 * ??? potentially stale state information?
938 */
939 if (part->sn.uv.remote_act_state != XPC_P_AS_DEACTIVATING &&
940 part->sn.uv.remote_act_state != XPC_P_AS_INACTIVE) {
941
942 msg.reason = part->reason;
943 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
944 XPC_ACTIVATE_MQ_MSG_DEACTIVATE_REQ_UV);
945 }
946 }
947
948 static void
949 xpc_cancel_partition_deactivation_request_uv(struct xpc_partition *part)
950 {
951 /* nothing needs to be done */
952 return;
953 }
954
955 static void
956 xpc_init_fifo_uv(struct xpc_fifo_head_uv *head)
957 {
958 head->first = NULL;
959 head->last = NULL;
960 spin_lock_init(&head->lock);
961 head->n_entries = 0;
962 }
963
964 static void *
965 xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
966 {
967 unsigned long irq_flags;
968 struct xpc_fifo_entry_uv *first;
969
970 spin_lock_irqsave(&head->lock, irq_flags);
971 first = head->first;
972 if (head->first != NULL) {
973 head->first = first->next;
974 if (head->first == NULL)
975 head->last = NULL;
976
977 head->n_entries--;
978 BUG_ON(head->n_entries < 0);
979
980 first->next = NULL;
981 }
982 spin_unlock_irqrestore(&head->lock, irq_flags);
983 return first;
984 }
985
986 static void
987 xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
988 struct xpc_fifo_entry_uv *last)
989 {
990 unsigned long irq_flags;
991
992 last->next = NULL;
993 spin_lock_irqsave(&head->lock, irq_flags);
994 if (head->last != NULL)
995 head->last->next = last;
996 else
997 head->first = last;
998 head->last = last;
999 head->n_entries++;
1000 spin_unlock_irqrestore(&head->lock, irq_flags);
1001 }
1002
1003 static int
1004 xpc_n_of_fifo_entries_uv(struct xpc_fifo_head_uv *head)
1005 {
1006 return head->n_entries;
1007 }
1008
1009 /*
1010 * Setup the channel structures that are uv specific.
1011 */
1012 static enum xp_retval
1013 xpc_setup_ch_structures_uv(struct xpc_partition *part)
1014 {
1015 struct xpc_channel_uv *ch_uv;
1016 int ch_number;
1017
1018 for (ch_number = 0; ch_number < part->nchannels; ch_number++) {
1019 ch_uv = &part->channels[ch_number].sn.uv;
1020
1021 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1022 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1023 }
1024
1025 return xpSuccess;
1026 }
1027
1028 /*
1029 * Teardown the channel structures that are uv specific.
1030 */
1031 static void
1032 xpc_teardown_ch_structures_uv(struct xpc_partition *part)
1033 {
1034 /* nothing needs to be done */
1035 return;
1036 }
1037
1038 static enum xp_retval
1039 xpc_make_first_contact_uv(struct xpc_partition *part)
1040 {
1041 struct xpc_activate_mq_msg_uv msg;
1042
1043 /*
1044 * We send a sync msg to get the remote partition's remote_act_state
1045 * updated to our current act_state which at this point should
1046 * be XPC_P_AS_ACTIVATING.
1047 */
1048 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1049 XPC_ACTIVATE_MQ_MSG_SYNC_ACT_STATE_UV);
1050
1051 while (!((part->sn.uv.remote_act_state == XPC_P_AS_ACTIVATING) ||
1052 (part->sn.uv.remote_act_state == XPC_P_AS_ACTIVE))) {
1053
1054 dev_dbg(xpc_part, "waiting to make first contact with "
1055 "partition %d\n", XPC_PARTID(part));
1056
1057 /* wait a 1/4 of a second or so */
1058 (void)msleep_interruptible(250);
1059
1060 if (part->act_state == XPC_P_AS_DEACTIVATING)
1061 return part->reason;
1062 }
1063
1064 return xpSuccess;
1065 }
1066
1067 static u64
1068 xpc_get_chctl_all_flags_uv(struct xpc_partition *part)
1069 {
1070 unsigned long irq_flags;
1071 union xpc_channel_ctl_flags chctl;
1072
1073 spin_lock_irqsave(&part->chctl_lock, irq_flags);
1074 chctl = part->chctl;
1075 if (chctl.all_flags != 0)
1076 part->chctl.all_flags = 0;
1077
1078 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1079 return chctl.all_flags;
1080 }
1081
1082 static enum xp_retval
1083 xpc_allocate_send_msg_slot_uv(struct xpc_channel *ch)
1084 {
1085 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1086 struct xpc_send_msg_slot_uv *msg_slot;
1087 unsigned long irq_flags;
1088 int nentries;
1089 int entry;
1090 size_t nbytes;
1091
1092 for (nentries = ch->local_nentries; nentries > 0; nentries--) {
1093 nbytes = nentries * sizeof(struct xpc_send_msg_slot_uv);
1094 ch_uv->send_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1095 if (ch_uv->send_msg_slots == NULL)
1096 continue;
1097
1098 for (entry = 0; entry < nentries; entry++) {
1099 msg_slot = &ch_uv->send_msg_slots[entry];
1100
1101 msg_slot->msg_slot_number = entry;
1102 xpc_put_fifo_entry_uv(&ch_uv->msg_slot_free_list,
1103 &msg_slot->next);
1104 }
1105
1106 spin_lock_irqsave(&ch->lock, irq_flags);
1107 if (nentries < ch->local_nentries)
1108 ch->local_nentries = nentries;
1109 spin_unlock_irqrestore(&ch->lock, irq_flags);
1110 return xpSuccess;
1111 }
1112
1113 return xpNoMemory;
1114 }
1115
1116 static enum xp_retval
1117 xpc_allocate_recv_msg_slot_uv(struct xpc_channel *ch)
1118 {
1119 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1120 struct xpc_notify_mq_msg_uv *msg_slot;
1121 unsigned long irq_flags;
1122 int nentries;
1123 int entry;
1124 size_t nbytes;
1125
1126 for (nentries = ch->remote_nentries; nentries > 0; nentries--) {
1127 nbytes = nentries * ch->entry_size;
1128 ch_uv->recv_msg_slots = kzalloc(nbytes, GFP_KERNEL);
1129 if (ch_uv->recv_msg_slots == NULL)
1130 continue;
1131
1132 for (entry = 0; entry < nentries; entry++) {
1133 msg_slot = ch_uv->recv_msg_slots +
1134 entry * ch->entry_size;
1135
1136 msg_slot->hdr.msg_slot_number = entry;
1137 }
1138
1139 spin_lock_irqsave(&ch->lock, irq_flags);
1140 if (nentries < ch->remote_nentries)
1141 ch->remote_nentries = nentries;
1142 spin_unlock_irqrestore(&ch->lock, irq_flags);
1143 return xpSuccess;
1144 }
1145
1146 return xpNoMemory;
1147 }
1148
1149 /*
1150 * Allocate msg_slots associated with the channel.
1151 */
1152 static enum xp_retval
1153 xpc_setup_msg_structures_uv(struct xpc_channel *ch)
1154 {
1155 static enum xp_retval ret;
1156 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1157
1158 DBUG_ON(ch->flags & XPC_C_SETUP);
1159
1160 ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
1161 gru_message_queue_desc),
1162 GFP_KERNEL);
1163 if (ch_uv->cached_notify_gru_mq_desc == NULL)
1164 return xpNoMemory;
1165
1166 ret = xpc_allocate_send_msg_slot_uv(ch);
1167 if (ret == xpSuccess) {
1168
1169 ret = xpc_allocate_recv_msg_slot_uv(ch);
1170 if (ret != xpSuccess) {
1171 kfree(ch_uv->send_msg_slots);
1172 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1173 }
1174 }
1175 return ret;
1176 }
1177
1178 /*
1179 * Free up msg_slots and clear other stuff that were setup for the specified
1180 * channel.
1181 */
1182 static void
1183 xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
1184 {
1185 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1186
1187 lockdep_assert_held(&ch->lock);
1188
1189 kfree(ch_uv->cached_notify_gru_mq_desc);
1190 ch_uv->cached_notify_gru_mq_desc = NULL;
1191
1192 if (ch->flags & XPC_C_SETUP) {
1193 xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
1194 kfree(ch_uv->send_msg_slots);
1195 xpc_init_fifo_uv(&ch_uv->recv_msg_list);
1196 kfree(ch_uv->recv_msg_slots);
1197 }
1198 }
1199
1200 static void
1201 xpc_send_chctl_closerequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1202 {
1203 struct xpc_activate_mq_msg_chctl_closerequest_uv msg;
1204
1205 msg.ch_number = ch->number;
1206 msg.reason = ch->reason;
1207 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1208 XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREQUEST_UV);
1209 }
1210
1211 static void
1212 xpc_send_chctl_closereply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1213 {
1214 struct xpc_activate_mq_msg_chctl_closereply_uv msg;
1215
1216 msg.ch_number = ch->number;
1217 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1218 XPC_ACTIVATE_MQ_MSG_CHCTL_CLOSEREPLY_UV);
1219 }
1220
1221 static void
1222 xpc_send_chctl_openrequest_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1223 {
1224 struct xpc_activate_mq_msg_chctl_openrequest_uv msg;
1225
1226 msg.ch_number = ch->number;
1227 msg.entry_size = ch->entry_size;
1228 msg.local_nentries = ch->local_nentries;
1229 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1230 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREQUEST_UV);
1231 }
1232
1233 static void
1234 xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1235 {
1236 struct xpc_activate_mq_msg_chctl_openreply_uv msg;
1237
1238 msg.ch_number = ch->number;
1239 msg.local_nentries = ch->local_nentries;
1240 msg.remote_nentries = ch->remote_nentries;
1241 msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
1242 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1243 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
1244 }
1245
1246 static void
1247 xpc_send_chctl_opencomplete_uv(struct xpc_channel *ch, unsigned long *irq_flags)
1248 {
1249 struct xpc_activate_mq_msg_chctl_opencomplete_uv msg;
1250
1251 msg.ch_number = ch->number;
1252 xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
1253 XPC_ACTIVATE_MQ_MSG_CHCTL_OPENCOMPLETE_UV);
1254 }
1255
1256 static void
1257 xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
1258 {
1259 unsigned long irq_flags;
1260
1261 spin_lock_irqsave(&part->chctl_lock, irq_flags);
1262 part->chctl.flags[ch_number] |= XPC_CHCTL_MSGREQUEST;
1263 spin_unlock_irqrestore(&part->chctl_lock, irq_flags);
1264
1265 xpc_wakeup_channel_mgr(part);
1266 }
1267
1268 static enum xp_retval
1269 xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
1270 unsigned long gru_mq_desc_gpa)
1271 {
1272 struct xpc_channel_uv *ch_uv = &ch->sn.uv;
1273
1274 DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
1275 return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
1276 gru_mq_desc_gpa);
1277 }
1278
1279 static void
1280 xpc_indicate_partition_engaged_uv(struct xpc_partition *part)
1281 {
1282 struct xpc_activate_mq_msg_uv msg;
1283
1284 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1285 XPC_ACTIVATE_MQ_MSG_MARK_ENGAGED_UV);
1286 }
1287
1288 static void
1289 xpc_indicate_partition_disengaged_uv(struct xpc_partition *part)
1290 {
1291 struct xpc_activate_mq_msg_uv msg;
1292
1293 xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
1294 XPC_ACTIVATE_MQ_MSG_MARK_DISENGAGED_UV);
1295 }
1296
1297 static void
1298 xpc_assume_partition_disengaged_uv(short partid)
1299 {
1300 struct xpc_partition_uv *part_uv = &xpc_partitions[partid].sn.uv;
1301 unsigned long irq_flags;
1302
1303 spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
1304 part_uv->flags &= ~XPC_P_ENGAGED_UV;
1305 spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
1306 }
1307
1308 static int
1309 xpc_partition_engaged_uv(short partid)
1310 {
1311 return (xpc_partitions[partid].sn.uv.flags & XPC_P_ENGAGED_UV) != 0;
1312 }
1313
1314 static int
1315 xpc_any_partition_engaged_uv(void)
1316 {
1317 struct xpc_partition_uv *part_uv;
1318 short partid;
1319
1320 for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
1321 part_uv = &xpc_partitions[partid].sn.uv;
1322 if ((part_uv->flags & XPC_P_ENGAGED_UV) != 0)
1323 return 1;
1324 }
1325 return 0;
1326 }
1327
1328 static enum xp_retval
1329 xpc_allocate_msg_slot_uv(struct xpc_channel *ch, u32 flags,
1330 struct xpc_send_msg_slot_uv **address_of_msg_slot)
1331 {
1332 enum xp_retval ret;
1333 struct xpc_send_msg_slot_uv *msg_slot;
1334 struct xpc_fifo_entry_uv *entry;
1335
1336 while (1) {
1337 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list);
1338 if (entry != NULL)
1339 break;
1340
1341 if (flags & XPC_NOWAIT)
1342 return xpNoWait;
1343
1344 ret = xpc_allocate_msg_wait(ch);
1345 if (ret != xpInterrupted && ret != xpTimeout)
1346 return ret;
1347 }
1348
1349 msg_slot = container_of(entry, struct xpc_send_msg_slot_uv, next);
1350 *address_of_msg_slot = msg_slot;
1351 return xpSuccess;
1352 }
1353
1354 static void
1355 xpc_free_msg_slot_uv(struct xpc_channel *ch,
1356 struct xpc_send_msg_slot_uv *msg_slot)
1357 {
1358 xpc_put_fifo_entry_uv(&ch->sn.uv.msg_slot_free_list, &msg_slot->next);
1359
1360 /* wakeup anyone waiting for a free msg slot */
1361 if (atomic_read(&ch->n_on_msg_allocate_wq) > 0)
1362 wake_up(&ch->msg_allocate_wq);
1363 }
1364
1365 static void
1366 xpc_notify_sender_uv(struct xpc_channel *ch,
1367 struct xpc_send_msg_slot_uv *msg_slot,
1368 enum xp_retval reason)
1369 {
1370 xpc_notify_func func = msg_slot->func;
1371
1372 if (func != NULL && cmpxchg(&msg_slot->func, func, NULL) == func) {
1373
1374 atomic_dec(&ch->n_to_notify);
1375
1376 dev_dbg(xpc_chan, "msg_slot->func() called, msg_slot=0x%p "
1377 "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1378 msg_slot->msg_slot_number, ch->partid, ch->number);
1379
1380 func(reason, ch->partid, ch->number, msg_slot->key);
1381
1382 dev_dbg(xpc_chan, "msg_slot->func() returned, msg_slot=0x%p "
1383 "msg_slot_number=%d partid=%d channel=%d\n", msg_slot,
1384 msg_slot->msg_slot_number, ch->partid, ch->number);
1385 }
1386 }
1387
1388 static void
1389 xpc_handle_notify_mq_ack_uv(struct xpc_channel *ch,
1390 struct xpc_notify_mq_msg_uv *msg)
1391 {
1392 struct xpc_send_msg_slot_uv *msg_slot;
1393 int entry = msg->hdr.msg_slot_number % ch->local_nentries;
1394
1395 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1396
1397 BUG_ON(msg_slot->msg_slot_number != msg->hdr.msg_slot_number);
1398 msg_slot->msg_slot_number += ch->local_nentries;
1399
1400 if (msg_slot->func != NULL)
1401 xpc_notify_sender_uv(ch, msg_slot, xpMsgDelivered);
1402
1403 xpc_free_msg_slot_uv(ch, msg_slot);
1404 }
1405
1406 static void
1407 xpc_handle_notify_mq_msg_uv(struct xpc_partition *part,
1408 struct xpc_notify_mq_msg_uv *msg)
1409 {
1410 struct xpc_partition_uv *part_uv = &part->sn.uv;
1411 struct xpc_channel *ch;
1412 struct xpc_channel_uv *ch_uv;
1413 struct xpc_notify_mq_msg_uv *msg_slot;
1414 unsigned long irq_flags;
1415 int ch_number = msg->hdr.ch_number;
1416
1417 if (unlikely(ch_number >= part->nchannels)) {
1418 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received invalid "
1419 "channel number=0x%x in message from partid=%d\n",
1420 ch_number, XPC_PARTID(part));
1421
1422 /* get hb checker to deactivate from the remote partition */
1423 spin_lock_irqsave(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1424 if (part_uv->act_state_req == 0)
1425 xpc_activate_IRQ_rcvd++;
1426 part_uv->act_state_req = XPC_P_ASR_DEACTIVATE_UV;
1427 part_uv->reason = xpBadChannelNumber;
1428 spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
1429
1430 wake_up_interruptible(&xpc_activate_IRQ_wq);
1431 return;
1432 }
1433
1434 ch = &part->channels[ch_number];
1435 xpc_msgqueue_ref(ch);
1436
1437 if (!(ch->flags & XPC_C_CONNECTED)) {
1438 xpc_msgqueue_deref(ch);
1439 return;
1440 }
1441
1442 /* see if we're really dealing with an ACK for a previously sent msg */
1443 if (msg->hdr.size == 0) {
1444 xpc_handle_notify_mq_ack_uv(ch, msg);
1445 xpc_msgqueue_deref(ch);
1446 return;
1447 }
1448
1449 /* we're dealing with a normal message sent via the notify_mq */
1450 ch_uv = &ch->sn.uv;
1451
1452 msg_slot = ch_uv->recv_msg_slots +
1453 (msg->hdr.msg_slot_number % ch->remote_nentries) * ch->entry_size;
1454
1455 BUG_ON(msg_slot->hdr.size != 0);
1456
1457 memcpy(msg_slot, msg, msg->hdr.size);
1458
1459 xpc_put_fifo_entry_uv(&ch_uv->recv_msg_list, &msg_slot->hdr.u.next);
1460
1461 if (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE) {
1462 /*
1463 * If there is an existing idle kthread get it to deliver
1464 * the payload, otherwise we'll have to get the channel mgr
1465 * for this partition to create a kthread to do the delivery.
1466 */
1467 if (atomic_read(&ch->kthreads_idle) > 0)
1468 wake_up_nr(&ch->idle_wq, 1);
1469 else
1470 xpc_send_chctl_local_msgrequest_uv(part, ch->number);
1471 }
1472 xpc_msgqueue_deref(ch);
1473 }
1474
1475 static irqreturn_t
1476 xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
1477 {
1478 struct xpc_notify_mq_msg_uv *msg;
1479 short partid;
1480 struct xpc_partition *part;
1481
1482 while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
1483 NULL) {
1484
1485 partid = msg->hdr.partid;
1486 if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
1487 dev_err(xpc_part, "xpc_handle_notify_IRQ_uv() received "
1488 "invalid partid=0x%x in message\n", partid);
1489 } else {
1490 part = &xpc_partitions[partid];
1491
1492 if (xpc_part_ref(part)) {
1493 xpc_handle_notify_mq_msg_uv(part, msg);
1494 xpc_part_deref(part);
1495 }
1496 }
1497
1498 gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
1499 }
1500
1501 return IRQ_HANDLED;
1502 }
1503
1504 static int
1505 xpc_n_of_deliverable_payloads_uv(struct xpc_channel *ch)
1506 {
1507 return xpc_n_of_fifo_entries_uv(&ch->sn.uv.recv_msg_list);
1508 }
1509
1510 static void
1511 xpc_process_msg_chctl_flags_uv(struct xpc_partition *part, int ch_number)
1512 {
1513 struct xpc_channel *ch = &part->channels[ch_number];
1514 int ndeliverable_payloads;
1515
1516 xpc_msgqueue_ref(ch);
1517
1518 ndeliverable_payloads = xpc_n_of_deliverable_payloads_uv(ch);
1519
1520 if (ndeliverable_payloads > 0 &&
1521 (ch->flags & XPC_C_CONNECTED) &&
1522 (ch->flags & XPC_C_CONNECTEDCALLOUT_MADE)) {
1523
1524 xpc_activate_kthreads(ch, ndeliverable_payloads);
1525 }
1526
1527 xpc_msgqueue_deref(ch);
1528 }
1529
1530 static enum xp_retval
1531 xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
1532 u16 payload_size, u8 notify_type, xpc_notify_func func,
1533 void *key)
1534 {
1535 enum xp_retval ret = xpSuccess;
1536 struct xpc_send_msg_slot_uv *msg_slot = NULL;
1537 struct xpc_notify_mq_msg_uv *msg;
1538 u8 msg_buffer[XPC_NOTIFY_MSG_SIZE_UV];
1539 size_t msg_size;
1540
1541 DBUG_ON(notify_type != XPC_N_CALL);
1542
1543 msg_size = sizeof(struct xpc_notify_mq_msghdr_uv) + payload_size;
1544 if (msg_size > ch->entry_size)
1545 return xpPayloadTooBig;
1546
1547 xpc_msgqueue_ref(ch);
1548
1549 if (ch->flags & XPC_C_DISCONNECTING) {
1550 ret = ch->reason;
1551 goto out_1;
1552 }
1553 if (!(ch->flags & XPC_C_CONNECTED)) {
1554 ret = xpNotConnected;
1555 goto out_1;
1556 }
1557
1558 ret = xpc_allocate_msg_slot_uv(ch, flags, &msg_slot);
1559 if (ret != xpSuccess)
1560 goto out_1;
1561
1562 if (func != NULL) {
1563 atomic_inc(&ch->n_to_notify);
1564
1565 msg_slot->key = key;
1566 smp_wmb(); /* a non-NULL func must hit memory after the key */
1567 msg_slot->func = func;
1568
1569 if (ch->flags & XPC_C_DISCONNECTING) {
1570 ret = ch->reason;
1571 goto out_2;
1572 }
1573 }
1574
1575 msg = (struct xpc_notify_mq_msg_uv *)&msg_buffer;
1576 msg->hdr.partid = xp_partition_id;
1577 msg->hdr.ch_number = ch->number;
1578 msg->hdr.size = msg_size;
1579 msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
1580 memcpy(&msg->payload, payload, payload_size);
1581
1582 ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1583 msg_size);
1584 if (ret == xpSuccess)
1585 goto out_1;
1586
1587 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1588 out_2:
1589 if (func != NULL) {
1590 /*
1591 * Try to NULL the msg_slot's func field. If we fail, then
1592 * xpc_notify_senders_of_disconnect_uv() beat us to it, in which
1593 * case we need to pretend we succeeded to send the message
1594 * since the user will get a callout for the disconnect error
1595 * by xpc_notify_senders_of_disconnect_uv(), and to also get an
1596 * error returned here will confuse them. Additionally, since
1597 * in this case the channel is being disconnected we don't need
1598 * to put the the msg_slot back on the free list.
1599 */
1600 if (cmpxchg(&msg_slot->func, func, NULL) != func) {
1601 ret = xpSuccess;
1602 goto out_1;
1603 }
1604
1605 msg_slot->key = NULL;
1606 atomic_dec(&ch->n_to_notify);
1607 }
1608 xpc_free_msg_slot_uv(ch, msg_slot);
1609 out_1:
1610 xpc_msgqueue_deref(ch);
1611 return ret;
1612 }
1613
1614 /*
1615 * Tell the callers of xpc_send_notify() that the status of their payloads
1616 * is unknown because the channel is now disconnecting.
1617 *
1618 * We don't worry about putting these msg_slots on the free list since the
1619 * msg_slots themselves are about to be kfree'd.
1620 */
1621 static void
1622 xpc_notify_senders_of_disconnect_uv(struct xpc_channel *ch)
1623 {
1624 struct xpc_send_msg_slot_uv *msg_slot;
1625 int entry;
1626
1627 DBUG_ON(!(ch->flags & XPC_C_DISCONNECTING));
1628
1629 for (entry = 0; entry < ch->local_nentries; entry++) {
1630
1631 if (atomic_read(&ch->n_to_notify) == 0)
1632 break;
1633
1634 msg_slot = &ch->sn.uv.send_msg_slots[entry];
1635 if (msg_slot->func != NULL)
1636 xpc_notify_sender_uv(ch, msg_slot, ch->reason);
1637 }
1638 }
1639
1640 /*
1641 * Get the next deliverable message's payload.
1642 */
1643 static void *
1644 xpc_get_deliverable_payload_uv(struct xpc_channel *ch)
1645 {
1646 struct xpc_fifo_entry_uv *entry;
1647 struct xpc_notify_mq_msg_uv *msg;
1648 void *payload = NULL;
1649
1650 if (!(ch->flags & XPC_C_DISCONNECTING)) {
1651 entry = xpc_get_fifo_entry_uv(&ch->sn.uv.recv_msg_list);
1652 if (entry != NULL) {
1653 msg = container_of(entry, struct xpc_notify_mq_msg_uv,
1654 hdr.u.next);
1655 payload = &msg->payload;
1656 }
1657 }
1658 return payload;
1659 }
1660
1661 static void
1662 xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
1663 {
1664 struct xpc_notify_mq_msg_uv *msg;
1665 enum xp_retval ret;
1666
1667 msg = container_of(payload, struct xpc_notify_mq_msg_uv, payload);
1668
1669 /* return an ACK to the sender of this message */
1670
1671 msg->hdr.partid = xp_partition_id;
1672 msg->hdr.size = 0; /* size of zero indicates this is an ACK */
1673
1674 ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
1675 sizeof(struct xpc_notify_mq_msghdr_uv));
1676 if (ret != xpSuccess)
1677 XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
1678 }
1679
1680 static struct xpc_arch_operations xpc_arch_ops_uv = {
1681 .setup_partitions = xpc_setup_partitions_uv,
1682 .teardown_partitions = xpc_teardown_partitions_uv,
1683 .process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv,
1684 .get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv,
1685 .setup_rsvd_page = xpc_setup_rsvd_page_uv,
1686
1687 .allow_hb = xpc_allow_hb_uv,
1688 .disallow_hb = xpc_disallow_hb_uv,
1689 .disallow_all_hbs = xpc_disallow_all_hbs_uv,
1690 .increment_heartbeat = xpc_increment_heartbeat_uv,
1691 .offline_heartbeat = xpc_offline_heartbeat_uv,
1692 .online_heartbeat = xpc_online_heartbeat_uv,
1693 .heartbeat_init = xpc_heartbeat_init_uv,
1694 .heartbeat_exit = xpc_heartbeat_exit_uv,
1695 .get_remote_heartbeat = xpc_get_remote_heartbeat_uv,
1696
1697 .request_partition_activation =
1698 xpc_request_partition_activation_uv,
1699 .request_partition_reactivation =
1700 xpc_request_partition_reactivation_uv,
1701 .request_partition_deactivation =
1702 xpc_request_partition_deactivation_uv,
1703 .cancel_partition_deactivation_request =
1704 xpc_cancel_partition_deactivation_request_uv,
1705
1706 .setup_ch_structures = xpc_setup_ch_structures_uv,
1707 .teardown_ch_structures = xpc_teardown_ch_structures_uv,
1708
1709 .make_first_contact = xpc_make_first_contact_uv,
1710
1711 .get_chctl_all_flags = xpc_get_chctl_all_flags_uv,
1712 .send_chctl_closerequest = xpc_send_chctl_closerequest_uv,
1713 .send_chctl_closereply = xpc_send_chctl_closereply_uv,
1714 .send_chctl_openrequest = xpc_send_chctl_openrequest_uv,
1715 .send_chctl_openreply = xpc_send_chctl_openreply_uv,
1716 .send_chctl_opencomplete = xpc_send_chctl_opencomplete_uv,
1717 .process_msg_chctl_flags = xpc_process_msg_chctl_flags_uv,
1718
1719 .save_remote_msgqueue_pa = xpc_save_remote_msgqueue_pa_uv,
1720
1721 .setup_msg_structures = xpc_setup_msg_structures_uv,
1722 .teardown_msg_structures = xpc_teardown_msg_structures_uv,
1723
1724 .indicate_partition_engaged = xpc_indicate_partition_engaged_uv,
1725 .indicate_partition_disengaged = xpc_indicate_partition_disengaged_uv,
1726 .assume_partition_disengaged = xpc_assume_partition_disengaged_uv,
1727 .partition_engaged = xpc_partition_engaged_uv,
1728 .any_partition_engaged = xpc_any_partition_engaged_uv,
1729
1730 .n_of_deliverable_payloads = xpc_n_of_deliverable_payloads_uv,
1731 .send_payload = xpc_send_payload_uv,
1732 .get_deliverable_payload = xpc_get_deliverable_payload_uv,
1733 .received_payload = xpc_received_payload_uv,
1734 .notify_senders_of_disconnect = xpc_notify_senders_of_disconnect_uv,
1735 };
1736
1737 static int
1738 xpc_init_mq_node(int nid)
1739 {
1740 int cpu;
1741
1742 get_online_cpus();
1743
1744 for_each_cpu(cpu, cpumask_of_node(nid)) {
1745 xpc_activate_mq_uv =
1746 xpc_create_gru_mq_uv(XPC_ACTIVATE_MQ_SIZE_UV, nid,
1747 XPC_ACTIVATE_IRQ_NAME,
1748 xpc_handle_activate_IRQ_uv);
1749 if (!IS_ERR(xpc_activate_mq_uv))
1750 break;
1751 }
1752 if (IS_ERR(xpc_activate_mq_uv)) {
1753 put_online_cpus();
1754 return PTR_ERR(xpc_activate_mq_uv);
1755 }
1756
1757 for_each_cpu(cpu, cpumask_of_node(nid)) {
1758 xpc_notify_mq_uv =
1759 xpc_create_gru_mq_uv(XPC_NOTIFY_MQ_SIZE_UV, nid,
1760 XPC_NOTIFY_IRQ_NAME,
1761 xpc_handle_notify_IRQ_uv);
1762 if (!IS_ERR(xpc_notify_mq_uv))
1763 break;
1764 }
1765 if (IS_ERR(xpc_notify_mq_uv)) {
1766 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1767 put_online_cpus();
1768 return PTR_ERR(xpc_notify_mq_uv);
1769 }
1770
1771 put_online_cpus();
1772 return 0;
1773 }
1774
1775 int
1776 xpc_init_uv(void)
1777 {
1778 int nid;
1779 int ret = 0;
1780
1781 xpc_arch_ops = xpc_arch_ops_uv;
1782
1783 if (sizeof(struct xpc_notify_mq_msghdr_uv) > XPC_MSG_HDR_MAX_SIZE) {
1784 dev_err(xpc_part, "xpc_notify_mq_msghdr_uv is larger than %d\n",
1785 XPC_MSG_HDR_MAX_SIZE);
1786 return -E2BIG;
1787 }
1788
1789 if (xpc_mq_node < 0)
1790 for_each_online_node(nid) {
1791 ret = xpc_init_mq_node(nid);
1792
1793 if (!ret)
1794 break;
1795 }
1796 else
1797 ret = xpc_init_mq_node(xpc_mq_node);
1798
1799 if (ret < 0)
1800 dev_err(xpc_part, "xpc_init_mq_node() returned error=%d\n",
1801 -ret);
1802
1803 return ret;
1804 }
1805
1806 void
1807 xpc_exit_uv(void)
1808 {
1809 xpc_destroy_gru_mq_uv(xpc_notify_mq_uv);
1810 xpc_destroy_gru_mq_uv(xpc_activate_mq_uv);
1811 }
1812
1813 module_param(xpc_mq_node, int, 0);
1814 MODULE_PARM_DESC(xpc_mq_node, "Node number on which to allocate message queues.");
Ubuntu Jammy Linux kernel mirror