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
6 * Copyright (c) 2004-2008 Silicon Graphics, Inc. All Rights Reserved.
10 * Cross Partition Communication (XPC) partition support.
12 * This is the part of XPC that detects the presence/absence of
13 * other partitions. It provides a heartbeat and monitors the
14 * heartbeats of other partitions.
18 #include <linux/device.h>
19 #include <linux/hardirq.h>
22 /* XPC is exiting flag */
25 /* this partition's reserved page pointers */
26 struct xpc_rsvd_page *xpc_rsvd_page;
27 static unsigned long *xpc_part_nasids;
28 unsigned long *xpc_mach_nasids;
30 static int xpc_nasid_mask_nbytes; /* #of bytes in nasid mask */
31 int xpc_nasid_mask_nlongs; /* #of longs in nasid mask */
33 struct xpc_partition *xpc_partitions;
36 * Guarantee that the kmalloc'd memory is cacheline aligned.
39 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
41 /* see if kmalloc will give us cachline aligned memory by default */
42 *base = kmalloc(size, flags);
46 if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
51 /* nope, we'll have to do it ourselves */
52 *base = kmalloc(size + L1_CACHE_BYTES, flags);
56 return (void *)L1_CACHE_ALIGN((u64)*base);
60 * Given a nasid, get the physical address of the partition's reserved page
61 * for that nasid. This function returns 0 on any error.
64 xpc_get_rsvd_page_pa(int nasid)
68 u64 rp_pa = nasid; /* seed with nasid */
72 void *buf_base = NULL;
76 ret = xpc_get_partition_rsvd_page_pa(buf, &cookie, &rp_pa,
79 dev_dbg(xpc_part, "SAL returned with ret=%d, cookie=0x%016lx, "
80 "address=0x%016lx, len=0x%016lx\n", ret,
81 (unsigned long)cookie, (unsigned long)rp_pa, len);
83 if (ret != xpNeedMoreInfo)
86 /* !!! L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
87 if (L1_CACHE_ALIGN(len) > buf_len) {
89 buf_len = L1_CACHE_ALIGN(len);
90 buf = (u64)xpc_kmalloc_cacheline_aligned(buf_len,
93 if (buf_base == NULL) {
94 dev_err(xpc_part, "unable to kmalloc "
96 (unsigned long)buf_len);
102 ret = xp_remote_memcpy((void *)buf, (void *)rp_pa, buf_len);
103 if (ret != xpSuccess) {
104 dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
111 if (ret != xpSuccess)
114 dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n",
115 (unsigned long)rp_pa);
120 * Fill the partition reserved page with the information needed by
121 * other partitions to discover we are alive and establish initial
124 struct xpc_rsvd_page *
125 xpc_setup_rsvd_page(void)
127 struct xpc_rsvd_page *rp;
129 unsigned long new_ts_jiffies;
131 /* get the local reserved page's address */
134 rp_pa = xpc_get_rsvd_page_pa(xp_cpu_to_nasid(smp_processor_id()));
137 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
140 rp = (struct xpc_rsvd_page *)__va(rp_pa);
142 if (rp->SAL_version < 3) {
143 /* SAL_versions < 3 had a SAL_partid defined as a u8 */
144 rp->SAL_partid &= 0xff;
146 BUG_ON(rp->SAL_partid != xp_partition_id);
148 if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
149 dev_err(xpc_part, "the reserved page's partid of %d is outside "
150 "supported range (< 0 || >= %d)\n", rp->SAL_partid,
155 rp->version = XPC_RP_VERSION;
156 rp->max_npartitions = xp_max_npartitions;
158 /* establish the actual sizes of the nasid masks */
159 if (rp->SAL_version == 1) {
160 /* SAL_version 1 didn't set the nasids_size field */
161 rp->SAL_nasids_size = 128;
163 xpc_nasid_mask_nbytes = rp->SAL_nasids_size;
164 xpc_nasid_mask_nlongs = BITS_TO_LONGS(rp->SAL_nasids_size *
167 /* setup the pointers to the various items in the reserved page */
168 xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
169 xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
171 if (xpc_rsvd_page_init(rp) != xpSuccess)
175 * Set timestamp of when reserved page was setup by XPC.
176 * This signifies to the remote partition that our reserved
177 * page is initialized.
179 new_ts_jiffies = jiffies;
180 if (new_ts_jiffies == 0 || new_ts_jiffies == rp->ts_jiffies)
182 rp->ts_jiffies = new_ts_jiffies;
188 * Get a copy of a portion of the remote partition's rsvd page.
190 * remote_rp points to a buffer that is cacheline aligned for BTE copies and
191 * is large enough to contain a copy of their reserved page header and
195 xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,
196 struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
201 /* get the reserved page's physical address */
203 *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
204 if (*remote_rp_pa == 0)
205 return xpNoRsvdPageAddr;
207 /* pull over the reserved page header and part_nasids mask */
208 ret = xp_remote_memcpy(remote_rp, (void *)*remote_rp_pa,
209 XPC_RP_HEADER_SIZE + xpc_nasid_mask_nbytes);
210 if (ret != xpSuccess)
213 if (discovered_nasids != NULL) {
214 unsigned long *remote_part_nasids =
215 XPC_RP_PART_NASIDS(remote_rp);
217 for (l = 0; l < xpc_nasid_mask_nlongs; l++)
218 discovered_nasids[l] |= remote_part_nasids[l];
221 /* zero timestamp indicates the reserved page has not been setup */
222 if (remote_rp->ts_jiffies == 0)
223 return xpRsvdPageNotSet;
225 if (XPC_VERSION_MAJOR(remote_rp->version) !=
226 XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
230 /* check that both remote and local partids are valid for each side */
231 if (remote_rp->SAL_partid < 0 ||
232 remote_rp->SAL_partid >= xp_max_npartitions ||
233 remote_rp->max_npartitions <= xp_partition_id) {
234 return xpInvalidPartid;
237 if (remote_rp->SAL_partid == xp_partition_id)
238 return xpLocalPartid;
244 * See if the other side has responded to a partition deactivate request
245 * from us. Though we requested the remote partition to deactivate with regard
246 * to us, we really only need to wait for the other side to disengage from us.
249 xpc_partition_disengaged(struct xpc_partition *part)
251 short partid = XPC_PARTID(part);
254 disengaged = !xpc_partition_engaged(partid);
255 if (part->disengage_timeout) {
257 if (time_is_after_jiffies(part->disengage_timeout)) {
258 /* timelimit hasn't been reached yet */
263 * Other side hasn't responded to our deactivate
264 * request in a timely fashion, so assume it's dead.
267 dev_info(xpc_part, "deactivate request to remote "
268 "partition %d timed out\n", partid);
269 xpc_disengage_timedout = 1;
270 xpc_assume_partition_disengaged(partid);
273 part->disengage_timeout = 0;
275 /* cancel the timer function, provided it's not us */
277 del_singleshot_timer_sync(&part->disengage_timer);
279 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
280 part->act_state != XPC_P_INACTIVE);
281 if (part->act_state != XPC_P_INACTIVE)
282 xpc_wakeup_channel_mgr(part);
284 xpc_cancel_partition_deactivation_request(part);
290 * Mark specified partition as active.
293 xpc_mark_partition_active(struct xpc_partition *part)
295 unsigned long irq_flags;
298 dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
300 spin_lock_irqsave(&part->act_lock, irq_flags);
301 if (part->act_state == XPC_P_ACTIVATING) {
302 part->act_state = XPC_P_ACTIVE;
305 DBUG_ON(part->reason == xpSuccess);
308 spin_unlock_irqrestore(&part->act_lock, irq_flags);
314 * Start the process of deactivating the specified partition.
317 xpc_deactivate_partition(const int line, struct xpc_partition *part,
318 enum xp_retval reason)
320 unsigned long irq_flags;
322 spin_lock_irqsave(&part->act_lock, irq_flags);
324 if (part->act_state == XPC_P_INACTIVE) {
325 XPC_SET_REASON(part, reason, line);
326 spin_unlock_irqrestore(&part->act_lock, irq_flags);
327 if (reason == xpReactivating) {
328 /* we interrupt ourselves to reactivate partition */
329 xpc_request_partition_reactivation(part);
333 if (part->act_state == XPC_P_DEACTIVATING) {
334 if ((part->reason == xpUnloading && reason != xpUnloading) ||
335 reason == xpReactivating) {
336 XPC_SET_REASON(part, reason, line);
338 spin_unlock_irqrestore(&part->act_lock, irq_flags);
342 part->act_state = XPC_P_DEACTIVATING;
343 XPC_SET_REASON(part, reason, line);
345 spin_unlock_irqrestore(&part->act_lock, irq_flags);
347 /* ask remote partition to deactivate with regard to us */
348 xpc_request_partition_deactivation(part);
350 /* set a timelimit on the disengage phase of the deactivation request */
351 part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
352 part->disengage_timer.expires = part->disengage_timeout;
353 add_timer(&part->disengage_timer);
355 dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
356 XPC_PARTID(part), reason);
358 xpc_partition_going_down(part, reason);
362 * Mark specified partition as inactive.
365 xpc_mark_partition_inactive(struct xpc_partition *part)
367 unsigned long irq_flags;
369 dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
372 spin_lock_irqsave(&part->act_lock, irq_flags);
373 part->act_state = XPC_P_INACTIVE;
374 spin_unlock_irqrestore(&part->act_lock, irq_flags);
375 part->remote_rp_pa = 0;
379 * SAL has provided a partition and machine mask. The partition mask
380 * contains a bit for each even nasid in our partition. The machine
381 * mask contains a bit for each even nasid in the entire machine.
383 * Using those two bit arrays, we can determine which nasids are
384 * known in the machine. Each should also have a reserved page
385 * initialized if they are available for partitioning.
390 void *remote_rp_base;
391 struct xpc_rsvd_page *remote_rp;
397 struct xpc_rsvd_page *rp;
398 unsigned long *discovered_nasids;
401 remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
402 xpc_nasid_mask_nbytes,
403 GFP_KERNEL, &remote_rp_base);
404 if (remote_rp == NULL)
407 discovered_nasids = kzalloc(sizeof(long) * xpc_nasid_mask_nlongs,
409 if (discovered_nasids == NULL) {
410 kfree(remote_rp_base);
414 rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
417 * The term 'region' in this context refers to the minimum number of
418 * nodes that can comprise an access protection grouping. The access
419 * protection is in regards to memory, IOI and IPI.
422 region_size = xp_region_size;
424 switch (region_size) {
432 DBUG_ON(!is_shub2());
435 for (region = 0; region < max_regions; region++) {
440 dev_dbg(xpc_part, "searching region %d\n", region);
442 for (nasid = (region * region_size * 2);
443 nasid < ((region + 1) * region_size * 2); nasid += 2) {
448 dev_dbg(xpc_part, "checking nasid %d\n", nasid);
450 if (test_bit(nasid / 2, xpc_part_nasids)) {
451 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
452 "part of the local partition; skipping "
457 if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
458 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
459 "not on Numa-Link network at reset\n",
464 if (test_bit(nasid / 2, discovered_nasids)) {
465 dev_dbg(xpc_part, "Nasid %d is part of a "
466 "partition which was previously "
467 "discovered\n", nasid);
471 /* pull over the rsvd page header & part_nasids mask */
473 ret = xpc_get_remote_rp(nasid, discovered_nasids,
474 remote_rp, &remote_rp_pa);
475 if (ret != xpSuccess) {
476 dev_dbg(xpc_part, "unable to get reserved page "
477 "from nasid %d, reason=%d\n", nasid,
480 if (ret == xpLocalPartid)
486 xpc_request_partition_activation(remote_rp,
487 remote_rp_pa, nasid);
491 kfree(discovered_nasids);
492 kfree(remote_rp_base);
496 * Given a partid, get the nasids owned by that partition from the
497 * remote partition's reserved page.
500 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
502 struct xpc_partition *part;
505 part = &xpc_partitions[partid];
506 if (part->remote_rp_pa == 0)
507 return xpPartitionDown;
509 memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
511 part_nasid_pa = (u64)XPC_RP_PART_NASIDS(part->remote_rp_pa);
513 return xp_remote_memcpy(nasid_mask, (void *)part_nasid_pa,
514 xpc_nasid_mask_nbytes);