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/kernel.h>
19 #include <linux/sysctl.h>
20 #include <linux/cache.h>
21 #include <linux/mmzone.h>
22 #include <linux/nodemask.h>
23 #include <asm/sn/intr.h>
24 #include <asm/sn/sn_sal.h>
25 #include <asm/sn/nodepda.h>
26 #include <asm/sn/addrs.h>
29 /* XPC is exiting flag */
32 /* this partition's reserved page pointers */
33 struct xpc_rsvd_page *xpc_rsvd_page;
34 static unsigned long *xpc_part_nasids;
35 unsigned long *xpc_mach_nasids;
37 static int xpc_nasid_mask_nbytes; /* #of bytes in nasid mask */
38 int xpc_nasid_mask_nlongs; /* #of longs in nasid mask */
40 struct xpc_partition *xpc_partitions;
43 * Guarantee that the kmalloc'd memory is cacheline aligned.
46 xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
48 /* see if kmalloc will give us cachline aligned memory by default */
49 *base = kmalloc(size, flags);
53 if ((u64)*base == L1_CACHE_ALIGN((u64)*base))
58 /* nope, we'll have to do it ourselves */
59 *base = kmalloc(size + L1_CACHE_BYTES, flags);
63 return (void *)L1_CACHE_ALIGN((u64)*base);
67 * Given a nasid, get the physical address of the partition's reserved page
68 * for that nasid. This function returns 0 on any error.
71 xpc_get_rsvd_page_pa(int nasid)
76 u64 rp_pa = nasid; /* seed with nasid */
80 void *buf_base = NULL;
84 status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
87 dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
88 "0x%016lx, address=0x%016lx, len=0x%016lx\n",
89 status, cookie, rp_pa, len);
91 if (status != SALRET_MORE_PASSES)
94 /* !!! L1_CACHE_ALIGN() is only a sn2-bte_copy requirement */
95 if (L1_CACHE_ALIGN(len) > buf_len) {
97 buf_len = L1_CACHE_ALIGN(len);
98 buf = (u64)xpc_kmalloc_cacheline_aligned(buf_len,
101 if (buf_base == NULL) {
102 dev_err(xpc_part, "unable to kmalloc "
103 "len=0x%016lx\n", buf_len);
104 status = SALRET_ERROR;
109 ret = xp_remote_memcpy((void *)buf, (void *)rp_pa, buf_len);
110 if (ret != xpSuccess) {
111 dev_dbg(xpc_part, "xp_remote_memcpy failed %d\n", ret);
112 status = SALRET_ERROR;
119 if (status != SALRET_OK)
122 dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
127 * Fill the partition reserved page with the information needed by
128 * other partitions to discover we are alive and establish initial
131 struct xpc_rsvd_page *
132 xpc_setup_rsvd_page(void)
134 struct xpc_rsvd_page *rp;
136 unsigned long new_ts_jiffies;
138 /* get the local reserved page's address */
141 rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
144 dev_err(xpc_part, "SAL failed to locate the reserved page\n");
147 rp = (struct xpc_rsvd_page *)__va(rp_pa);
149 if (rp->SAL_version < 3) {
150 /* SAL_versions < 3 had a SAL_partid defined as a u8 */
151 rp->SAL_partid &= 0xff;
153 BUG_ON(rp->SAL_partid != sn_partition_id);
155 if (rp->SAL_partid < 0 || rp->SAL_partid >= xp_max_npartitions) {
156 dev_err(xpc_part, "the reserved page's partid of %d is outside "
157 "supported range (< 0 || >= %d)\n", rp->SAL_partid,
162 rp->version = XPC_RP_VERSION;
163 rp->max_npartitions = xp_max_npartitions;
165 /* establish the actual sizes of the nasid masks */
166 if (rp->SAL_version == 1) {
167 /* SAL_version 1 didn't set the nasids_size field */
168 rp->SAL_nasids_size = 128;
170 xpc_nasid_mask_nbytes = rp->SAL_nasids_size;
171 xpc_nasid_mask_nlongs = BITS_TO_LONGS(rp->SAL_nasids_size *
174 /* setup the pointers to the various items in the reserved page */
175 xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
176 xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
178 if (xpc_rsvd_page_init(rp) != xpSuccess)
182 * Set timestamp of when reserved page was setup by XPC.
183 * This signifies to the remote partition that our reserved
184 * page is initialized.
186 new_ts_jiffies = jiffies;
187 if (new_ts_jiffies == 0 || new_ts_jiffies == rp->ts_jiffies)
189 rp->ts_jiffies = new_ts_jiffies;
195 * Get a copy of a portion of the remote partition's rsvd page.
197 * remote_rp points to a buffer that is cacheline aligned for BTE copies and
198 * is large enough to contain a copy of their reserved page header and
202 xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,
203 struct xpc_rsvd_page *remote_rp, u64 *remote_rp_pa)
208 /* get the reserved page's physical address */
210 *remote_rp_pa = xpc_get_rsvd_page_pa(nasid);
211 if (*remote_rp_pa == 0)
212 return xpNoRsvdPageAddr;
214 /* pull over the reserved page header and part_nasids mask */
215 ret = xp_remote_memcpy(remote_rp, (void *)*remote_rp_pa,
216 XPC_RP_HEADER_SIZE + xpc_nasid_mask_nbytes);
217 if (ret != xpSuccess)
220 if (discovered_nasids != NULL) {
221 unsigned long *remote_part_nasids =
222 XPC_RP_PART_NASIDS(remote_rp);
224 for (l = 0; l < xpc_nasid_mask_nlongs; l++)
225 discovered_nasids[l] |= remote_part_nasids[l];
228 /* zero timestamp indicates the reserved page has not been setup */
229 if (remote_rp->ts_jiffies == 0)
230 return xpRsvdPageNotSet;
232 if (XPC_VERSION_MAJOR(remote_rp->version) !=
233 XPC_VERSION_MAJOR(XPC_RP_VERSION)) {
237 /* check that both remote and local partids are valid for each side */
238 if (remote_rp->SAL_partid < 0 ||
239 remote_rp->SAL_partid >= xp_max_npartitions ||
240 remote_rp->max_npartitions <= sn_partition_id) {
241 return xpInvalidPartid;
244 if (remote_rp->SAL_partid == sn_partition_id)
245 return xpLocalPartid;
251 * See if the other side has responded to a partition deactivate request
252 * from us. Though we requested the remote partition to deactivate with regard
253 * to us, we really only need to wait for the other side to disengage from us.
256 xpc_partition_disengaged(struct xpc_partition *part)
258 short partid = XPC_PARTID(part);
261 disengaged = !xpc_partition_engaged(partid);
262 if (part->disengage_timeout) {
264 if (time_is_after_jiffies(part->disengage_timeout)) {
265 /* timelimit hasn't been reached yet */
270 * Other side hasn't responded to our deactivate
271 * request in a timely fashion, so assume it's dead.
274 dev_info(xpc_part, "deactivate request to remote "
275 "partition %d timed out\n", partid);
276 xpc_disengage_timedout = 1;
277 xpc_assume_partition_disengaged(partid);
280 part->disengage_timeout = 0;
282 /* cancel the timer function, provided it's not us */
284 del_singleshot_timer_sync(&part->disengage_timer);
286 DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
287 part->act_state != XPC_P_INACTIVE);
288 if (part->act_state != XPC_P_INACTIVE)
289 xpc_wakeup_channel_mgr(part);
291 xpc_cancel_partition_deactivation_request(part);
297 * Mark specified partition as active.
300 xpc_mark_partition_active(struct xpc_partition *part)
302 unsigned long irq_flags;
305 dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));
307 spin_lock_irqsave(&part->act_lock, irq_flags);
308 if (part->act_state == XPC_P_ACTIVATING) {
309 part->act_state = XPC_P_ACTIVE;
312 DBUG_ON(part->reason == xpSuccess);
315 spin_unlock_irqrestore(&part->act_lock, irq_flags);
321 * Start the process of deactivating the specified partition.
324 xpc_deactivate_partition(const int line, struct xpc_partition *part,
325 enum xp_retval reason)
327 unsigned long irq_flags;
329 spin_lock_irqsave(&part->act_lock, irq_flags);
331 if (part->act_state == XPC_P_INACTIVE) {
332 XPC_SET_REASON(part, reason, line);
333 spin_unlock_irqrestore(&part->act_lock, irq_flags);
334 if (reason == xpReactivating) {
335 /* we interrupt ourselves to reactivate partition */
336 xpc_request_partition_reactivation(part);
340 if (part->act_state == XPC_P_DEACTIVATING) {
341 if ((part->reason == xpUnloading && reason != xpUnloading) ||
342 reason == xpReactivating) {
343 XPC_SET_REASON(part, reason, line);
345 spin_unlock_irqrestore(&part->act_lock, irq_flags);
349 part->act_state = XPC_P_DEACTIVATING;
350 XPC_SET_REASON(part, reason, line);
352 spin_unlock_irqrestore(&part->act_lock, irq_flags);
354 /* ask remote partition to deactivate with regard to us */
355 xpc_request_partition_deactivation(part);
357 /* set a timelimit on the disengage phase of the deactivation request */
358 part->disengage_timeout = jiffies + (xpc_disengage_timelimit * HZ);
359 part->disengage_timer.expires = part->disengage_timeout;
360 add_timer(&part->disengage_timer);
362 dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
363 XPC_PARTID(part), reason);
365 xpc_partition_going_down(part, reason);
369 * Mark specified partition as inactive.
372 xpc_mark_partition_inactive(struct xpc_partition *part)
374 unsigned long irq_flags;
376 dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
379 spin_lock_irqsave(&part->act_lock, irq_flags);
380 part->act_state = XPC_P_INACTIVE;
381 spin_unlock_irqrestore(&part->act_lock, irq_flags);
382 part->remote_rp_pa = 0;
386 * SAL has provided a partition and machine mask. The partition mask
387 * contains a bit for each even nasid in our partition. The machine
388 * mask contains a bit for each even nasid in the entire machine.
390 * Using those two bit arrays, we can determine which nasids are
391 * known in the machine. Each should also have a reserved page
392 * initialized if they are available for partitioning.
397 void *remote_rp_base;
398 struct xpc_rsvd_page *remote_rp;
404 struct xpc_rsvd_page *rp;
405 unsigned long *discovered_nasids;
408 remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
409 xpc_nasid_mask_nbytes,
410 GFP_KERNEL, &remote_rp_base);
411 if (remote_rp == NULL)
414 discovered_nasids = kzalloc(sizeof(long) * xpc_nasid_mask_nlongs,
416 if (discovered_nasids == NULL) {
417 kfree(remote_rp_base);
421 rp = (struct xpc_rsvd_page *)xpc_rsvd_page;
424 * The term 'region' in this context refers to the minimum number of
425 * nodes that can comprise an access protection grouping. The access
426 * protection is in regards to memory, IOI and IPI.
429 region_size = sn_region_size;
431 switch (region_size) {
439 DBUG_ON(!is_shub2());
442 for (region = 0; region < max_regions; region++) {
447 dev_dbg(xpc_part, "searching region %d\n", region);
449 for (nasid = (region * region_size * 2);
450 nasid < ((region + 1) * region_size * 2); nasid += 2) {
455 dev_dbg(xpc_part, "checking nasid %d\n", nasid);
457 if (test_bit(nasid / 2, xpc_part_nasids)) {
458 dev_dbg(xpc_part, "PROM indicates Nasid %d is "
459 "part of the local partition; skipping "
464 if (!(test_bit(nasid / 2, xpc_mach_nasids))) {
465 dev_dbg(xpc_part, "PROM indicates Nasid %d was "
466 "not on Numa-Link network at reset\n",
471 if (test_bit(nasid / 2, discovered_nasids)) {
472 dev_dbg(xpc_part, "Nasid %d is part of a "
473 "partition which was previously "
474 "discovered\n", nasid);
478 /* pull over the rsvd page header & part_nasids mask */
480 ret = xpc_get_remote_rp(nasid, discovered_nasids,
481 remote_rp, &remote_rp_pa);
482 if (ret != xpSuccess) {
483 dev_dbg(xpc_part, "unable to get reserved page "
484 "from nasid %d, reason=%d\n", nasid,
487 if (ret == xpLocalPartid)
493 xpc_request_partition_activation(remote_rp,
494 remote_rp_pa, nasid);
498 kfree(discovered_nasids);
499 kfree(remote_rp_base);
503 * Given a partid, get the nasids owned by that partition from the
504 * remote partition's reserved page.
507 xpc_initiate_partid_to_nasids(short partid, void *nasid_mask)
509 struct xpc_partition *part;
512 part = &xpc_partitions[partid];
513 if (part->remote_rp_pa == 0)
514 return xpPartitionDown;
516 memset(nasid_mask, 0, xpc_nasid_mask_nbytes);
518 part_nasid_pa = (u64)XPC_RP_PART_NASIDS(part->remote_rp_pa);
520 return xp_remote_memcpy(nasid_mask, (void *)part_nasid_pa,
521 xpc_nasid_mask_nbytes);