aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/rcutree.c
diff options
context:
space:
mode:
authorPaul E. McKenney <paulmck@linux.vnet.ibm.com>2008-12-18 15:55:32 -0500
committerIngo Molnar <mingo@elte.hu>2008-12-18 15:56:04 -0500
commit64db4cfff99c04cd5f550357edcc8780f96b54a2 (patch)
tree4856e788d21f0e31ed78a22b70b4521f7237705e /kernel/rcutree.c
parentd110ec3a1e1f522e2e9dfceb9c36d6590c26d2d4 (diff)
"Tree RCU": scalable classic RCU implementation
This patch fixes a long-standing performance bug in classic RCU that results in massive internal-to-RCU lock contention on systems with more than a few hundred CPUs. Although this patch creates a separate flavor of RCU for ease of review and patch maintenance, it is intended to replace classic RCU. This patch still handles stress better than does mainline, so I am still calling it ready for inclusion. This patch is against the -tip tree. Nevertheless, experience on an actual 1000+ CPU machine would still be most welcome. Most of the changes noted below were found while creating an rcutiny (which should permit ejecting the current rcuclassic) and while doing detailed line-by-line documentation. Updates from v9 (http://lkml.org/lkml/2008/12/2/334): o Fixes from remainder of line-by-line code walkthrough, including comment spelling, initialization, undesirable narrowing due to type conversion, removing redundant memory barriers, removing redundant local-variable initialization, and removing redundant local variables. I do not believe that any of these fixes address the CPU-hotplug issues that Andi Kleen was seeing, but please do give it a whirl in case the machine is smarter than I am. A writeup from the walkthrough may be found at the following URL, in case you are suffering from terminal insomnia or masochism: http://www.kernel.org/pub/linux/kernel/people/paulmck/tmp/rcutree-walkthrough.2008.12.16a.pdf o Made rcutree tracing use seq_file, as suggested some time ago by Lai Jiangshan. o Added a .csv variant of the rcudata debugfs trace file, to allow people having thousands of CPUs to drop the data into a spreadsheet. Tested with oocalc and gnumeric. Updated documentation to suit. Updates from v8 (http://lkml.org/lkml/2008/11/15/139): o Fix a theoretical race between grace-period initialization and force_quiescent_state() that could occur if more than three jiffies were required to carry out the grace-period initialization. Which it might, if you had enough CPUs. o Apply Ingo's printk-standardization patch. o Substitute local variables for repeated accesses to global variables. o Fix comment misspellings and redundant (but harmless) increments of ->n_rcu_pending (this latter after having explicitly added it). o Apply checkpatch fixes. Updates from v7 (http://lkml.org/lkml/2008/10/10/291): o Fixed a number of problems noted by Gautham Shenoy, including the cpu-stall-detection bug that he was having difficulty convincing me was real. ;-) o Changed cpu-stall detection to wait for ten seconds rather than three in order to reduce false positive, as suggested by Ingo Molnar. o Produced a design document (http://lwn.net/Articles/305782/). The act of writing this document uncovered a number of both theoretical and "here and now" bugs as noted below. o Fix dynticks_nesting accounting confusion, simplify WARN_ON() condition, fix kerneldoc comments, and add memory barriers in dynticks interface functions. o Add more data to tracing. o Remove unused "rcu_barrier" field from rcu_data structure. o Count calls to rcu_pending() from scheduling-clock interrupt to use as a surrogate timebase should jiffies stop counting. o Fix a theoretical race between force_quiescent_state() and grace-period initialization. Yes, initialization does have to go on for some jiffies for this race to occur, but given enough CPUs... Updates from v6 (http://lkml.org/lkml/2008/9/23/448): o Fix a number of checkpatch.pl complaints. o Apply review comments from Ingo Molnar and Lai Jiangshan on the stall-detection code. o Fix several bugs in !CONFIG_SMP builds. o Fix a misspelled config-parameter name so that RCU now announces at boot time if stall detection is configured. o Run tests on numerous combinations of configurations parameters, which after the fixes above, now build and run correctly. Updates from v5 (http://lkml.org/lkml/2008/9/15/92, bad subject line): o Fix a compiler error in the !CONFIG_FANOUT_EXACT case (blew a changeset some time ago, and finally got around to retesting this option). o Fix some tracing bugs in rcupreempt that caused incorrect totals to be printed. o I now test with a more brutal random-selection online/offline script (attached). Probably more brutal than it needs to be on the people reading it as well, but so it goes. o A number of optimizations and usability improvements: o Make rcu_pending() ignore the grace-period timeout when there is no grace period in progress. o Make force_quiescent_state() avoid going for a global lock in the case where there is no grace period in progress. o Rearrange struct fields to improve struct layout. o Make call_rcu() initiate a grace period if RCU was idle, rather than waiting for the next scheduling clock interrupt. o Invoke rcu_irq_enter() and rcu_irq_exit() only when idle, as suggested by Andi Kleen. I still don't completely trust this change, and might back it out. o Make CONFIG_RCU_TRACE be the single config variable manipulated for all forms of RCU, instead of the prior confusion. o Document tracing files and formats for both rcupreempt and rcutree. Updates from v4 for those missing v5 given its bad subject line: o Separated dynticks interface so that NMIs and irqs call separate functions, greatly simplifying it. In particular, this code no longer requires a proof of correctness. ;-) o Separated dynticks state out into its own per-CPU structure, avoiding the duplicated accounting. o The case where a dynticks-idle CPU runs an irq handler that invokes call_rcu() is now correctly handled, forcing that CPU out of dynticks-idle mode. o Review comments have been applied (thank you all!!!). For but one example, fixed the dynticks-ordering issue that Manfred pointed out, saving me much debugging. ;-) o Adjusted rcuclassic and rcupreempt to handle dynticks changes. Attached is an updated patch to Classic RCU that applies a hierarchy, greatly reducing the contention on the top-level lock for large machines. This passes 10-hour concurrent rcutorture and online-offline testing on 128-CPU ppc64 without dynticks enabled, and exposes some timekeeping bugs in presence of dynticks (exciting working on a system where "sleep 1" hangs until interrupted...), which were fixed in the 2.6.27 kernel. It is getting more reliable than mainline by some measures, so the next version will be against -tip for inclusion. See also Manfred Spraul's recent patches (or his earlier work from 2004 at http://marc.info/?l=linux-kernel&m=108546384711797&w=2). We will converge onto a common patch in the fullness of time, but are currently exploring different regions of the design space. That said, I have already gratefully stolen quite a few of Manfred's ideas. This patch provides CONFIG_RCU_FANOUT, which controls the bushiness of the RCU hierarchy. Defaults to 32 on 32-bit machines and 64 on 64-bit machines. If CONFIG_NR_CPUS is less than CONFIG_RCU_FANOUT, there is no hierarchy. By default, the RCU initialization code will adjust CONFIG_RCU_FANOUT to balance the hierarchy, so strongly NUMA architectures may choose to set CONFIG_RCU_FANOUT_EXACT to disable this balancing, allowing the hierarchy to be exactly aligned to the underlying hardware. Up to two levels of hierarchy are permitted (in addition to the root node), allowing up to 16,384 CPUs on 32-bit systems and up to 262,144 CPUs on 64-bit systems. I just know that I am going to regret saying this, but this seems more than sufficient for the foreseeable future. (Some architectures might wish to set CONFIG_RCU_FANOUT=4, which would limit such architectures to 64 CPUs. If this becomes a real problem, additional levels can be added, but I doubt that it will make a significant difference on real hardware.) In the common case, a given CPU will manipulate its private rcu_data structure and the rcu_node structure that it shares with its immediate neighbors. This can reduce both lock and memory contention by multiple orders of magnitude, which should eliminate the need for the strange manipulations that are reported to be required when running Linux on very large systems. Some shortcomings: o More bugs will probably surface as a result of an ongoing line-by-line code inspection. Patches will be provided as required. o There are probably hangs, rcutorture failures, &c. Seems quite stable on a 128-CPU machine, but that is kind of small compared to 4096 CPUs. However, seems to do better than mainline. Patches will be provided as required. o The memory footprint of this version is several KB larger than rcuclassic. A separate UP-only rcutiny patch will be provided, which will reduce the memory footprint significantly, even compared to the old rcuclassic. One such patch passes light testing, and has a memory footprint smaller even than rcuclassic. Initial reaction from various embedded guys was "it is not worth it", so am putting it aside. Credits: o Manfred Spraul for ideas, review comments, and bugs spotted, as well as some good friendly competition. ;-) o Josh Triplett, Ingo Molnar, Peter Zijlstra, Mathieu Desnoyers, Lai Jiangshan, Andi Kleen, Andy Whitcroft, and Andrew Morton for reviews and comments. o Thomas Gleixner for much-needed help with some timer issues (see patches below). o Jon M. Tollefson, Tim Pepper, Andrew Theurer, Jose R. Santos, Andy Whitcroft, Darrick Wong, Nishanth Aravamudan, Anton Blanchard, Dave Kleikamp, and Nathan Lynch for keeping machines alive despite my heavy abuse^Wtesting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel/rcutree.c')
-rw-r--r--kernel/rcutree.c1535
1 files changed, 1535 insertions, 0 deletions
diff --git a/kernel/rcutree.c b/kernel/rcutree.c
new file mode 100644
index 000000000000..a342b032112c
--- /dev/null
+++ b/kernel/rcutree.c
@@ -0,0 +1,1535 @@
1/*
2 * Read-Copy Update mechanism for mutual exclusion
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * Copyright IBM Corporation, 2008
19 *
20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
21 * Manfred Spraul <manfred@colorfullife.com>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
23 *
24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
26 *
27 * For detailed explanation of Read-Copy Update mechanism see -
28 * Documentation/RCU
29 */
30#include <linux/types.h>
31#include <linux/kernel.h>
32#include <linux/init.h>
33#include <linux/spinlock.h>
34#include <linux/smp.h>
35#include <linux/rcupdate.h>
36#include <linux/interrupt.h>
37#include <linux/sched.h>
38#include <asm/atomic.h>
39#include <linux/bitops.h>
40#include <linux/module.h>
41#include <linux/completion.h>
42#include <linux/moduleparam.h>
43#include <linux/percpu.h>
44#include <linux/notifier.h>
45#include <linux/cpu.h>
46#include <linux/mutex.h>
47#include <linux/time.h>
48
49#ifdef CONFIG_DEBUG_LOCK_ALLOC
50static struct lock_class_key rcu_lock_key;
51struct lockdep_map rcu_lock_map =
52 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
53EXPORT_SYMBOL_GPL(rcu_lock_map);
54#endif
55
56/* Data structures. */
57
58#define RCU_STATE_INITIALIZER(name) { \
59 .level = { &name.node[0] }, \
60 .levelcnt = { \
61 NUM_RCU_LVL_0, /* root of hierarchy. */ \
62 NUM_RCU_LVL_1, \
63 NUM_RCU_LVL_2, \
64 NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \
65 }, \
66 .signaled = RCU_SIGNAL_INIT, \
67 .gpnum = -300, \
68 .completed = -300, \
69 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
70 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
71 .n_force_qs = 0, \
72 .n_force_qs_ngp = 0, \
73}
74
75struct rcu_state rcu_state = RCU_STATE_INITIALIZER(rcu_state);
76DEFINE_PER_CPU(struct rcu_data, rcu_data);
77
78struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
79DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
80
81#ifdef CONFIG_NO_HZ
82DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks);
83#endif /* #ifdef CONFIG_NO_HZ */
84
85static int blimit = 10; /* Maximum callbacks per softirq. */
86static int qhimark = 10000; /* If this many pending, ignore blimit. */
87static int qlowmark = 100; /* Once only this many pending, use blimit. */
88
89static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
90
91/*
92 * Return the number of RCU batches processed thus far for debug & stats.
93 */
94long rcu_batches_completed(void)
95{
96 return rcu_state.completed;
97}
98EXPORT_SYMBOL_GPL(rcu_batches_completed);
99
100/*
101 * Return the number of RCU BH batches processed thus far for debug & stats.
102 */
103long rcu_batches_completed_bh(void)
104{
105 return rcu_bh_state.completed;
106}
107EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
108
109/*
110 * Does the CPU have callbacks ready to be invoked?
111 */
112static int
113cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
114{
115 return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
116}
117
118/*
119 * Does the current CPU require a yet-as-unscheduled grace period?
120 */
121static int
122cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
123{
124 /* ACCESS_ONCE() because we are accessing outside of lock. */
125 return *rdp->nxttail[RCU_DONE_TAIL] &&
126 ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
127}
128
129/*
130 * Return the root node of the specified rcu_state structure.
131 */
132static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
133{
134 return &rsp->node[0];
135}
136
137#ifdef CONFIG_SMP
138
139/*
140 * If the specified CPU is offline, tell the caller that it is in
141 * a quiescent state. Otherwise, whack it with a reschedule IPI.
142 * Grace periods can end up waiting on an offline CPU when that
143 * CPU is in the process of coming online -- it will be added to the
144 * rcu_node bitmasks before it actually makes it online. The same thing
145 * can happen while a CPU is in the process of coming online. Because this
146 * race is quite rare, we check for it after detecting that the grace
147 * period has been delayed rather than checking each and every CPU
148 * each and every time we start a new grace period.
149 */
150static int rcu_implicit_offline_qs(struct rcu_data *rdp)
151{
152 /*
153 * If the CPU is offline, it is in a quiescent state. We can
154 * trust its state not to change because interrupts are disabled.
155 */
156 if (cpu_is_offline(rdp->cpu)) {
157 rdp->offline_fqs++;
158 return 1;
159 }
160
161 /* The CPU is online, so send it a reschedule IPI. */
162 if (rdp->cpu != smp_processor_id())
163 smp_send_reschedule(rdp->cpu);
164 else
165 set_need_resched();
166 rdp->resched_ipi++;
167 return 0;
168}
169
170#endif /* #ifdef CONFIG_SMP */
171
172#ifdef CONFIG_NO_HZ
173static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5);
174
175/**
176 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
177 *
178 * Enter nohz mode, in other words, -leave- the mode in which RCU
179 * read-side critical sections can occur. (Though RCU read-side
180 * critical sections can occur in irq handlers in nohz mode, a possibility
181 * handled by rcu_irq_enter() and rcu_irq_exit()).
182 */
183void rcu_enter_nohz(void)
184{
185 unsigned long flags;
186 struct rcu_dynticks *rdtp;
187
188 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
189 local_irq_save(flags);
190 rdtp = &__get_cpu_var(rcu_dynticks);
191 rdtp->dynticks++;
192 rdtp->dynticks_nesting--;
193 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
194 local_irq_restore(flags);
195}
196
197/*
198 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
199 *
200 * Exit nohz mode, in other words, -enter- the mode in which RCU
201 * read-side critical sections normally occur.
202 */
203void rcu_exit_nohz(void)
204{
205 unsigned long flags;
206 struct rcu_dynticks *rdtp;
207
208 local_irq_save(flags);
209 rdtp = &__get_cpu_var(rcu_dynticks);
210 rdtp->dynticks++;
211 rdtp->dynticks_nesting++;
212 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
213 local_irq_restore(flags);
214 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
215}
216
217/**
218 * rcu_nmi_enter - inform RCU of entry to NMI context
219 *
220 * If the CPU was idle with dynamic ticks active, and there is no
221 * irq handler running, this updates rdtp->dynticks_nmi to let the
222 * RCU grace-period handling know that the CPU is active.
223 */
224void rcu_nmi_enter(void)
225{
226 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
227
228 if (rdtp->dynticks & 0x1)
229 return;
230 rdtp->dynticks_nmi++;
231 WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs);
232 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
233}
234
235/**
236 * rcu_nmi_exit - inform RCU of exit from NMI context
237 *
238 * If the CPU was idle with dynamic ticks active, and there is no
239 * irq handler running, this updates rdtp->dynticks_nmi to let the
240 * RCU grace-period handling know that the CPU is no longer active.
241 */
242void rcu_nmi_exit(void)
243{
244 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
245
246 if (rdtp->dynticks & 0x1)
247 return;
248 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
249 rdtp->dynticks_nmi++;
250 WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs);
251}
252
253/**
254 * rcu_irq_enter - inform RCU of entry to hard irq context
255 *
256 * If the CPU was idle with dynamic ticks active, this updates the
257 * rdtp->dynticks to let the RCU handling know that the CPU is active.
258 */
259void rcu_irq_enter(void)
260{
261 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
262
263 if (rdtp->dynticks_nesting++)
264 return;
265 rdtp->dynticks++;
266 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs);
267 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
268}
269
270/**
271 * rcu_irq_exit - inform RCU of exit from hard irq context
272 *
273 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
274 * to put let the RCU handling be aware that the CPU is going back to idle
275 * with no ticks.
276 */
277void rcu_irq_exit(void)
278{
279 struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
280
281 if (--rdtp->dynticks_nesting)
282 return;
283 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
284 rdtp->dynticks++;
285 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs);
286
287 /* If the interrupt queued a callback, get out of dyntick mode. */
288 if (__get_cpu_var(rcu_data).nxtlist ||
289 __get_cpu_var(rcu_bh_data).nxtlist)
290 set_need_resched();
291}
292
293/*
294 * Record the specified "completed" value, which is later used to validate
295 * dynticks counter manipulations. Specify "rsp->completed - 1" to
296 * unconditionally invalidate any future dynticks manipulations (which is
297 * useful at the beginning of a grace period).
298 */
299static void dyntick_record_completed(struct rcu_state *rsp, long comp)
300{
301 rsp->dynticks_completed = comp;
302}
303
304#ifdef CONFIG_SMP
305
306/*
307 * Recall the previously recorded value of the completion for dynticks.
308 */
309static long dyntick_recall_completed(struct rcu_state *rsp)
310{
311 return rsp->dynticks_completed;
312}
313
314/*
315 * Snapshot the specified CPU's dynticks counter so that we can later
316 * credit them with an implicit quiescent state. Return 1 if this CPU
317 * is already in a quiescent state courtesy of dynticks idle mode.
318 */
319static int dyntick_save_progress_counter(struct rcu_data *rdp)
320{
321 int ret;
322 int snap;
323 int snap_nmi;
324
325 snap = rdp->dynticks->dynticks;
326 snap_nmi = rdp->dynticks->dynticks_nmi;
327 smp_mb(); /* Order sampling of snap with end of grace period. */
328 rdp->dynticks_snap = snap;
329 rdp->dynticks_nmi_snap = snap_nmi;
330 ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
331 if (ret)
332 rdp->dynticks_fqs++;
333 return ret;
334}
335
336/*
337 * Return true if the specified CPU has passed through a quiescent
338 * state by virtue of being in or having passed through an dynticks
339 * idle state since the last call to dyntick_save_progress_counter()
340 * for this same CPU.
341 */
342static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
343{
344 long curr;
345 long curr_nmi;
346 long snap;
347 long snap_nmi;
348
349 curr = rdp->dynticks->dynticks;
350 snap = rdp->dynticks_snap;
351 curr_nmi = rdp->dynticks->dynticks_nmi;
352 snap_nmi = rdp->dynticks_nmi_snap;
353 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
354
355 /*
356 * If the CPU passed through or entered a dynticks idle phase with
357 * no active irq/NMI handlers, then we can safely pretend that the CPU
358 * already acknowledged the request to pass through a quiescent
359 * state. Either way, that CPU cannot possibly be in an RCU
360 * read-side critical section that started before the beginning
361 * of the current RCU grace period.
362 */
363 if ((curr != snap || (curr & 0x1) == 0) &&
364 (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
365 rdp->dynticks_fqs++;
366 return 1;
367 }
368
369 /* Go check for the CPU being offline. */
370 return rcu_implicit_offline_qs(rdp);
371}
372
373#endif /* #ifdef CONFIG_SMP */
374
375#else /* #ifdef CONFIG_NO_HZ */
376
377static void dyntick_record_completed(struct rcu_state *rsp, long comp)
378{
379}
380
381#ifdef CONFIG_SMP
382
383/*
384 * If there are no dynticks, then the only way that a CPU can passively
385 * be in a quiescent state is to be offline. Unlike dynticks idle, which
386 * is a point in time during the prior (already finished) grace period,
387 * an offline CPU is always in a quiescent state, and thus can be
388 * unconditionally applied. So just return the current value of completed.
389 */
390static long dyntick_recall_completed(struct rcu_state *rsp)
391{
392 return rsp->completed;
393}
394
395static int dyntick_save_progress_counter(struct rcu_data *rdp)
396{
397 return 0;
398}
399
400static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
401{
402 return rcu_implicit_offline_qs(rdp);
403}
404
405#endif /* #ifdef CONFIG_SMP */
406
407#endif /* #else #ifdef CONFIG_NO_HZ */
408
409#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
410
411static void record_gp_stall_check_time(struct rcu_state *rsp)
412{
413 rsp->gp_start = jiffies;
414 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
415}
416
417static void print_other_cpu_stall(struct rcu_state *rsp)
418{
419 int cpu;
420 long delta;
421 unsigned long flags;
422 struct rcu_node *rnp = rcu_get_root(rsp);
423 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
424 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
425
426 /* Only let one CPU complain about others per time interval. */
427
428 spin_lock_irqsave(&rnp->lock, flags);
429 delta = jiffies - rsp->jiffies_stall;
430 if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) {
431 spin_unlock_irqrestore(&rnp->lock, flags);
432 return;
433 }
434 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
435 spin_unlock_irqrestore(&rnp->lock, flags);
436
437 /* OK, time to rat on our buddy... */
438
439 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
440 for (; rnp_cur < rnp_end; rnp_cur++) {
441 if (rnp_cur->qsmask == 0)
442 continue;
443 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
444 if (rnp_cur->qsmask & (1UL << cpu))
445 printk(" %d", rnp_cur->grplo + cpu);
446 }
447 printk(" (detected by %d, t=%ld jiffies)\n",
448 smp_processor_id(), (long)(jiffies - rsp->gp_start));
449 force_quiescent_state(rsp, 0); /* Kick them all. */
450}
451
452static void print_cpu_stall(struct rcu_state *rsp)
453{
454 unsigned long flags;
455 struct rcu_node *rnp = rcu_get_root(rsp);
456
457 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n",
458 smp_processor_id(), jiffies - rsp->gp_start);
459 dump_stack();
460 spin_lock_irqsave(&rnp->lock, flags);
461 if ((long)(jiffies - rsp->jiffies_stall) >= 0)
462 rsp->jiffies_stall =
463 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
464 spin_unlock_irqrestore(&rnp->lock, flags);
465 set_need_resched(); /* kick ourselves to get things going. */
466}
467
468static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
469{
470 long delta;
471 struct rcu_node *rnp;
472
473 delta = jiffies - rsp->jiffies_stall;
474 rnp = rdp->mynode;
475 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) {
476
477 /* We haven't checked in, so go dump stack. */
478 print_cpu_stall(rsp);
479
480 } else if (rsp->gpnum != rsp->completed &&
481 delta >= RCU_STALL_RAT_DELAY) {
482
483 /* They had two time units to dump stack, so complain. */
484 print_other_cpu_stall(rsp);
485 }
486}
487
488#else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
489
490static void record_gp_stall_check_time(struct rcu_state *rsp)
491{
492}
493
494static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
495{
496}
497
498#endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
499
500/*
501 * Update CPU-local rcu_data state to record the newly noticed grace period.
502 * This is used both when we started the grace period and when we notice
503 * that someone else started the grace period.
504 */
505static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
506{
507 rdp->qs_pending = 1;
508 rdp->passed_quiesc = 0;
509 rdp->gpnum = rsp->gpnum;
510 rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending +
511 RCU_JIFFIES_TILL_FORCE_QS;
512}
513
514/*
515 * Did someone else start a new RCU grace period start since we last
516 * checked? Update local state appropriately if so. Must be called
517 * on the CPU corresponding to rdp.
518 */
519static int
520check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
521{
522 unsigned long flags;
523 int ret = 0;
524
525 local_irq_save(flags);
526 if (rdp->gpnum != rsp->gpnum) {
527 note_new_gpnum(rsp, rdp);
528 ret = 1;
529 }
530 local_irq_restore(flags);
531 return ret;
532}
533
534/*
535 * Start a new RCU grace period if warranted, re-initializing the hierarchy
536 * in preparation for detecting the next grace period. The caller must hold
537 * the root node's ->lock, which is released before return. Hard irqs must
538 * be disabled.
539 */
540static void
541rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
542 __releases(rcu_get_root(rsp)->lock)
543{
544 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
545 struct rcu_node *rnp = rcu_get_root(rsp);
546 struct rcu_node *rnp_cur;
547 struct rcu_node *rnp_end;
548
549 if (!cpu_needs_another_gp(rsp, rdp)) {
550 spin_unlock_irqrestore(&rnp->lock, flags);
551 return;
552 }
553
554 /* Advance to a new grace period and initialize state. */
555 rsp->gpnum++;
556 rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
557 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
558 rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending +
559 RCU_JIFFIES_TILL_FORCE_QS;
560 record_gp_stall_check_time(rsp);
561 dyntick_record_completed(rsp, rsp->completed - 1);
562 note_new_gpnum(rsp, rdp);
563
564 /*
565 * Because we are first, we know that all our callbacks will
566 * be covered by this upcoming grace period, even the ones
567 * that were registered arbitrarily recently.
568 */
569 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
570 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
571
572 /* Special-case the common single-level case. */
573 if (NUM_RCU_NODES == 1) {
574 rnp->qsmask = rnp->qsmaskinit;
575 spin_unlock_irqrestore(&rnp->lock, flags);
576 return;
577 }
578
579 spin_unlock(&rnp->lock); /* leave irqs disabled. */
580
581
582 /* Exclude any concurrent CPU-hotplug operations. */
583 spin_lock(&rsp->onofflock); /* irqs already disabled. */
584
585 /*
586 * Set the quiescent-state-needed bits in all the non-leaf RCU
587 * nodes for all currently online CPUs. This operation relies
588 * on the layout of the hierarchy within the rsp->node[] array.
589 * Note that other CPUs will access only the leaves of the
590 * hierarchy, which still indicate that no grace period is in
591 * progress. In addition, we have excluded CPU-hotplug operations.
592 *
593 * We therefore do not need to hold any locks. Any required
594 * memory barriers will be supplied by the locks guarding the
595 * leaf rcu_nodes in the hierarchy.
596 */
597
598 rnp_end = rsp->level[NUM_RCU_LVLS - 1];
599 for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++)
600 rnp_cur->qsmask = rnp_cur->qsmaskinit;
601
602 /*
603 * Now set up the leaf nodes. Here we must be careful. First,
604 * we need to hold the lock in order to exclude other CPUs, which
605 * might be contending for the leaf nodes' locks. Second, as
606 * soon as we initialize a given leaf node, its CPUs might run
607 * up the rest of the hierarchy. We must therefore acquire locks
608 * for each node that we touch during this stage. (But we still
609 * are excluding CPU-hotplug operations.)
610 *
611 * Note that the grace period cannot complete until we finish
612 * the initialization process, as there will be at least one
613 * qsmask bit set in the root node until that time, namely the
614 * one corresponding to this CPU.
615 */
616 rnp_end = &rsp->node[NUM_RCU_NODES];
617 rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
618 for (; rnp_cur < rnp_end; rnp_cur++) {
619 spin_lock(&rnp_cur->lock); /* irqs already disabled. */
620 rnp_cur->qsmask = rnp_cur->qsmaskinit;
621 spin_unlock(&rnp_cur->lock); /* irqs already disabled. */
622 }
623
624 rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
625 spin_unlock_irqrestore(&rsp->onofflock, flags);
626}
627
628/*
629 * Advance this CPU's callbacks, but only if the current grace period
630 * has ended. This may be called only from the CPU to whom the rdp
631 * belongs.
632 */
633static void
634rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
635{
636 long completed_snap;
637 unsigned long flags;
638
639 local_irq_save(flags);
640 completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */
641
642 /* Did another grace period end? */
643 if (rdp->completed != completed_snap) {
644
645 /* Advance callbacks. No harm if list empty. */
646 rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
647 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
648 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
649
650 /* Remember that we saw this grace-period completion. */
651 rdp->completed = completed_snap;
652 }
653 local_irq_restore(flags);
654}
655
656/*
657 * Similar to cpu_quiet(), for which it is a helper function. Allows
658 * a group of CPUs to be quieted at one go, though all the CPUs in the
659 * group must be represented by the same leaf rcu_node structure.
660 * That structure's lock must be held upon entry, and it is released
661 * before return.
662 */
663static void
664cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
665 unsigned long flags)
666 __releases(rnp->lock)
667{
668 /* Walk up the rcu_node hierarchy. */
669 for (;;) {
670 if (!(rnp->qsmask & mask)) {
671
672 /* Our bit has already been cleared, so done. */
673 spin_unlock_irqrestore(&rnp->lock, flags);
674 return;
675 }
676 rnp->qsmask &= ~mask;
677 if (rnp->qsmask != 0) {
678
679 /* Other bits still set at this level, so done. */
680 spin_unlock_irqrestore(&rnp->lock, flags);
681 return;
682 }
683 mask = rnp->grpmask;
684 if (rnp->parent == NULL) {
685
686 /* No more levels. Exit loop holding root lock. */
687
688 break;
689 }
690 spin_unlock_irqrestore(&rnp->lock, flags);
691 rnp = rnp->parent;
692 spin_lock_irqsave(&rnp->lock, flags);
693 }
694
695 /*
696 * Get here if we are the last CPU to pass through a quiescent
697 * state for this grace period. Clean up and let rcu_start_gp()
698 * start up the next grace period if one is needed. Note that
699 * we still hold rnp->lock, as required by rcu_start_gp(), which
700 * will release it.
701 */
702 rsp->completed = rsp->gpnum;
703 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
704 rcu_start_gp(rsp, flags); /* releases rnp->lock. */
705}
706
707/*
708 * Record a quiescent state for the specified CPU, which must either be
709 * the current CPU or an offline CPU. The lastcomp argument is used to
710 * make sure we are still in the grace period of interest. We don't want
711 * to end the current grace period based on quiescent states detected in
712 * an earlier grace period!
713 */
714static void
715cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
716{
717 unsigned long flags;
718 unsigned long mask;
719 struct rcu_node *rnp;
720
721 rnp = rdp->mynode;
722 spin_lock_irqsave(&rnp->lock, flags);
723 if (lastcomp != ACCESS_ONCE(rsp->completed)) {
724
725 /*
726 * Someone beat us to it for this grace period, so leave.
727 * The race with GP start is resolved by the fact that we
728 * hold the leaf rcu_node lock, so that the per-CPU bits
729 * cannot yet be initialized -- so we would simply find our
730 * CPU's bit already cleared in cpu_quiet_msk() if this race
731 * occurred.
732 */
733 rdp->passed_quiesc = 0; /* try again later! */
734 spin_unlock_irqrestore(&rnp->lock, flags);
735 return;
736 }
737 mask = rdp->grpmask;
738 if ((rnp->qsmask & mask) == 0) {
739 spin_unlock_irqrestore(&rnp->lock, flags);
740 } else {
741 rdp->qs_pending = 0;
742
743 /*
744 * This GP can't end until cpu checks in, so all of our
745 * callbacks can be processed during the next GP.
746 */
747 rdp = rsp->rda[smp_processor_id()];
748 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
749
750 cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */
751 }
752}
753
754/*
755 * Check to see if there is a new grace period of which this CPU
756 * is not yet aware, and if so, set up local rcu_data state for it.
757 * Otherwise, see if this CPU has just passed through its first
758 * quiescent state for this grace period, and record that fact if so.
759 */
760static void
761rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
762{
763 /* If there is now a new grace period, record and return. */
764 if (check_for_new_grace_period(rsp, rdp))
765 return;
766
767 /*
768 * Does this CPU still need to do its part for current grace period?
769 * If no, return and let the other CPUs do their part as well.
770 */
771 if (!rdp->qs_pending)
772 return;
773
774 /*
775 * Was there a quiescent state since the beginning of the grace
776 * period? If no, then exit and wait for the next call.
777 */
778 if (!rdp->passed_quiesc)
779 return;
780
781 /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */
782 cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
783}
784
785#ifdef CONFIG_HOTPLUG_CPU
786
787/*
788 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
789 * and move all callbacks from the outgoing CPU to the current one.
790 */
791static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
792{
793 int i;
794 unsigned long flags;
795 long lastcomp;
796 unsigned long mask;
797 struct rcu_data *rdp = rsp->rda[cpu];
798 struct rcu_data *rdp_me;
799 struct rcu_node *rnp;
800
801 /* Exclude any attempts to start a new grace period. */
802 spin_lock_irqsave(&rsp->onofflock, flags);
803
804 /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
805 rnp = rdp->mynode;
806 mask = rdp->grpmask; /* rnp->grplo is constant. */
807 do {
808 spin_lock(&rnp->lock); /* irqs already disabled. */
809 rnp->qsmaskinit &= ~mask;
810 if (rnp->qsmaskinit != 0) {
811 spin_unlock(&rnp->lock); /* irqs already disabled. */
812 break;
813 }
814 mask = rnp->grpmask;
815 spin_unlock(&rnp->lock); /* irqs already disabled. */
816 rnp = rnp->parent;
817 } while (rnp != NULL);
818 lastcomp = rsp->completed;
819
820 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
821
822 /* Being offline is a quiescent state, so go record it. */
823 cpu_quiet(cpu, rsp, rdp, lastcomp);
824
825 /*
826 * Move callbacks from the outgoing CPU to the running CPU.
827 * Note that the outgoing CPU is now quiscent, so it is now
828 * (uncharacteristically) safe to access it rcu_data structure.
829 * Note also that we must carefully retain the order of the
830 * outgoing CPU's callbacks in order for rcu_barrier() to work
831 * correctly. Finally, note that we start all the callbacks
832 * afresh, even those that have passed through a grace period
833 * and are therefore ready to invoke. The theory is that hotplug
834 * events are rare, and that if they are frequent enough to
835 * indefinitely delay callbacks, you have far worse things to
836 * be worrying about.
837 */
838 rdp_me = rsp->rda[smp_processor_id()];
839 if (rdp->nxtlist != NULL) {
840 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
841 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
842 rdp->nxtlist = NULL;
843 for (i = 0; i < RCU_NEXT_SIZE; i++)
844 rdp->nxttail[i] = &rdp->nxtlist;
845 rdp_me->qlen += rdp->qlen;
846 rdp->qlen = 0;
847 }
848 local_irq_restore(flags);
849}
850
851/*
852 * Remove the specified CPU from the RCU hierarchy and move any pending
853 * callbacks that it might have to the current CPU. This code assumes
854 * that at least one CPU in the system will remain running at all times.
855 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
856 */
857static void rcu_offline_cpu(int cpu)
858{
859 __rcu_offline_cpu(cpu, &rcu_state);
860 __rcu_offline_cpu(cpu, &rcu_bh_state);
861}
862
863#else /* #ifdef CONFIG_HOTPLUG_CPU */
864
865static void rcu_offline_cpu(int cpu)
866{
867}
868
869#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
870
871/*
872 * Invoke any RCU callbacks that have made it to the end of their grace
873 * period. Thottle as specified by rdp->blimit.
874 */
875static void rcu_do_batch(struct rcu_data *rdp)
876{
877 unsigned long flags;
878 struct rcu_head *next, *list, **tail;
879 int count;
880
881 /* If no callbacks are ready, just return.*/
882 if (!cpu_has_callbacks_ready_to_invoke(rdp))
883 return;
884
885 /*
886 * Extract the list of ready callbacks, disabling to prevent
887 * races with call_rcu() from interrupt handlers.
888 */
889 local_irq_save(flags);
890 list = rdp->nxtlist;
891 rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
892 *rdp->nxttail[RCU_DONE_TAIL] = NULL;
893 tail = rdp->nxttail[RCU_DONE_TAIL];
894 for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
895 if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
896 rdp->nxttail[count] = &rdp->nxtlist;
897 local_irq_restore(flags);
898
899 /* Invoke callbacks. */
900 count = 0;
901 while (list) {
902 next = list->next;
903 prefetch(next);
904 list->func(list);
905 list = next;
906 if (++count >= rdp->blimit)
907 break;
908 }
909
910 local_irq_save(flags);
911
912 /* Update count, and requeue any remaining callbacks. */
913 rdp->qlen -= count;
914 if (list != NULL) {
915 *tail = rdp->nxtlist;
916 rdp->nxtlist = list;
917 for (count = 0; count < RCU_NEXT_SIZE; count++)
918 if (&rdp->nxtlist == rdp->nxttail[count])
919 rdp->nxttail[count] = tail;
920 else
921 break;
922 }
923
924 /* Reinstate batch limit if we have worked down the excess. */
925 if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
926 rdp->blimit = blimit;
927
928 local_irq_restore(flags);
929
930 /* Re-raise the RCU softirq if there are callbacks remaining. */
931 if (cpu_has_callbacks_ready_to_invoke(rdp))
932 raise_softirq(RCU_SOFTIRQ);
933}
934
935/*
936 * Check to see if this CPU is in a non-context-switch quiescent state
937 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
938 * Also schedule the RCU softirq handler.
939 *
940 * This function must be called with hardirqs disabled. It is normally
941 * invoked from the scheduling-clock interrupt. If rcu_pending returns
942 * false, there is no point in invoking rcu_check_callbacks().
943 */
944void rcu_check_callbacks(int cpu, int user)
945{
946 if (user ||
947 (idle_cpu(cpu) && !in_softirq() &&
948 hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
949
950 /*
951 * Get here if this CPU took its interrupt from user
952 * mode or from the idle loop, and if this is not a
953 * nested interrupt. In this case, the CPU is in
954 * a quiescent state, so count it.
955 *
956 * No memory barrier is required here because both
957 * rcu_qsctr_inc() and rcu_bh_qsctr_inc() reference
958 * only CPU-local variables that other CPUs neither
959 * access nor modify, at least not while the corresponding
960 * CPU is online.
961 */
962
963 rcu_qsctr_inc(cpu);
964 rcu_bh_qsctr_inc(cpu);
965
966 } else if (!in_softirq()) {
967
968 /*
969 * Get here if this CPU did not take its interrupt from
970 * softirq, in other words, if it is not interrupting
971 * a rcu_bh read-side critical section. This is an _bh
972 * critical section, so count it.
973 */
974
975 rcu_bh_qsctr_inc(cpu);
976 }
977 raise_softirq(RCU_SOFTIRQ);
978}
979
980#ifdef CONFIG_SMP
981
982/*
983 * Scan the leaf rcu_node structures, processing dyntick state for any that
984 * have not yet encountered a quiescent state, using the function specified.
985 * Returns 1 if the current grace period ends while scanning (possibly
986 * because we made it end).
987 */
988static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
989 int (*f)(struct rcu_data *))
990{
991 unsigned long bit;
992 int cpu;
993 unsigned long flags;
994 unsigned long mask;
995 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
996 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
997
998 for (; rnp_cur < rnp_end; rnp_cur++) {
999 mask = 0;
1000 spin_lock_irqsave(&rnp_cur->lock, flags);
1001 if (rsp->completed != lastcomp) {
1002 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1003 return 1;
1004 }
1005 if (rnp_cur->qsmask == 0) {
1006 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1007 continue;
1008 }
1009 cpu = rnp_cur->grplo;
1010 bit = 1;
1011 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) {
1012 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1013 mask |= bit;
1014 }
1015 if (mask != 0 && rsp->completed == lastcomp) {
1016
1017 /* cpu_quiet_msk() releases rnp_cur->lock. */
1018 cpu_quiet_msk(mask, rsp, rnp_cur, flags);
1019 continue;
1020 }
1021 spin_unlock_irqrestore(&rnp_cur->lock, flags);
1022 }
1023 return 0;
1024}
1025
1026/*
1027 * Force quiescent states on reluctant CPUs, and also detect which
1028 * CPUs are in dyntick-idle mode.
1029 */
1030static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1031{
1032 unsigned long flags;
1033 long lastcomp;
1034 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
1035 struct rcu_node *rnp = rcu_get_root(rsp);
1036 u8 signaled;
1037
1038 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum))
1039 return; /* No grace period in progress, nothing to force. */
1040 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1041 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
1042 return; /* Someone else is already on the job. */
1043 }
1044 if (relaxed &&
1045 (long)(rsp->jiffies_force_qs - jiffies) >= 0 &&
1046 (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) >= 0)
1047 goto unlock_ret; /* no emergency and done recently. */
1048 rsp->n_force_qs++;
1049 spin_lock(&rnp->lock);
1050 lastcomp = rsp->completed;
1051 signaled = rsp->signaled;
1052 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1053 rdp->n_rcu_pending_force_qs = rdp->n_rcu_pending +
1054 RCU_JIFFIES_TILL_FORCE_QS;
1055 if (lastcomp == rsp->gpnum) {
1056 rsp->n_force_qs_ngp++;
1057 spin_unlock(&rnp->lock);
1058 goto unlock_ret; /* no GP in progress, time updated. */
1059 }
1060 spin_unlock(&rnp->lock);
1061 switch (signaled) {
1062 case RCU_GP_INIT:
1063
1064 break; /* grace period still initializing, ignore. */
1065
1066 case RCU_SAVE_DYNTICK:
1067
1068 if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1069 break; /* So gcc recognizes the dead code. */
1070
1071 /* Record dyntick-idle state. */
1072 if (rcu_process_dyntick(rsp, lastcomp,
1073 dyntick_save_progress_counter))
1074 goto unlock_ret;
1075
1076 /* Update state, record completion counter. */
1077 spin_lock(&rnp->lock);
1078 if (lastcomp == rsp->completed) {
1079 rsp->signaled = RCU_FORCE_QS;
1080 dyntick_record_completed(rsp, lastcomp);
1081 }
1082 spin_unlock(&rnp->lock);
1083 break;
1084
1085 case RCU_FORCE_QS:
1086
1087 /* Check dyntick-idle state, send IPI to laggarts. */
1088 if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp),
1089 rcu_implicit_dynticks_qs))
1090 goto unlock_ret;
1091
1092 /* Leave state in case more forcing is required. */
1093
1094 break;
1095 }
1096unlock_ret:
1097 spin_unlock_irqrestore(&rsp->fqslock, flags);
1098}
1099
1100#else /* #ifdef CONFIG_SMP */
1101
1102static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1103{
1104 set_need_resched();
1105}
1106
1107#endif /* #else #ifdef CONFIG_SMP */
1108
1109/*
1110 * This does the RCU processing work from softirq context for the
1111 * specified rcu_state and rcu_data structures. This may be called
1112 * only from the CPU to whom the rdp belongs.
1113 */
1114static void
1115__rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1116{
1117 unsigned long flags;
1118
1119 /*
1120 * If an RCU GP has gone long enough, go check for dyntick
1121 * idle CPUs and, if needed, send resched IPIs.
1122 */
1123 if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0 ||
1124 (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) < 0)
1125 force_quiescent_state(rsp, 1);
1126
1127 /*
1128 * Advance callbacks in response to end of earlier grace
1129 * period that some other CPU ended.
1130 */
1131 rcu_process_gp_end(rsp, rdp);
1132
1133 /* Update RCU state based on any recent quiescent states. */
1134 rcu_check_quiescent_state(rsp, rdp);
1135
1136 /* Does this CPU require a not-yet-started grace period? */
1137 if (cpu_needs_another_gp(rsp, rdp)) {
1138 spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1139 rcu_start_gp(rsp, flags); /* releases above lock */
1140 }
1141
1142 /* If there are callbacks ready, invoke them. */
1143 rcu_do_batch(rdp);
1144}
1145
1146/*
1147 * Do softirq processing for the current CPU.
1148 */
1149static void rcu_process_callbacks(struct softirq_action *unused)
1150{
1151 /*
1152 * Memory references from any prior RCU read-side critical sections
1153 * executed by the interrupted code must be seen before any RCU
1154 * grace-period manipulations below.
1155 */
1156 smp_mb(); /* See above block comment. */
1157
1158 __rcu_process_callbacks(&rcu_state, &__get_cpu_var(rcu_data));
1159 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1160
1161 /*
1162 * Memory references from any later RCU read-side critical sections
1163 * executed by the interrupted code must be seen after any RCU
1164 * grace-period manipulations above.
1165 */
1166 smp_mb(); /* See above block comment. */
1167}
1168
1169static void
1170__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1171 struct rcu_state *rsp)
1172{
1173 unsigned long flags;
1174 struct rcu_data *rdp;
1175
1176 head->func = func;
1177 head->next = NULL;
1178
1179 smp_mb(); /* Ensure RCU update seen before callback registry. */
1180
1181 /*
1182 * Opportunistically note grace-period endings and beginnings.
1183 * Note that we might see a beginning right after we see an
1184 * end, but never vice versa, since this CPU has to pass through
1185 * a quiescent state betweentimes.
1186 */
1187 local_irq_save(flags);
1188 rdp = rsp->rda[smp_processor_id()];
1189 rcu_process_gp_end(rsp, rdp);
1190 check_for_new_grace_period(rsp, rdp);
1191
1192 /* Add the callback to our list. */
1193 *rdp->nxttail[RCU_NEXT_TAIL] = head;
1194 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1195
1196 /* Start a new grace period if one not already started. */
1197 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) {
1198 unsigned long nestflag;
1199 struct rcu_node *rnp_root = rcu_get_root(rsp);
1200
1201 spin_lock_irqsave(&rnp_root->lock, nestflag);
1202 rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */
1203 }
1204
1205 /* Force the grace period if too many callbacks or too long waiting. */
1206 if (unlikely(++rdp->qlen > qhimark)) {
1207 rdp->blimit = LONG_MAX;
1208 force_quiescent_state(rsp, 0);
1209 } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0 ||
1210 (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) < 0)
1211 force_quiescent_state(rsp, 1);
1212 local_irq_restore(flags);
1213}
1214
1215/*
1216 * Queue an RCU callback for invocation after a grace period.
1217 */
1218void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1219{
1220 __call_rcu(head, func, &rcu_state);
1221}
1222EXPORT_SYMBOL_GPL(call_rcu);
1223
1224/*
1225 * Queue an RCU for invocation after a quicker grace period.
1226 */
1227void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1228{
1229 __call_rcu(head, func, &rcu_bh_state);
1230}
1231EXPORT_SYMBOL_GPL(call_rcu_bh);
1232
1233/*
1234 * Check to see if there is any immediate RCU-related work to be done
1235 * by the current CPU, for the specified type of RCU, returning 1 if so.
1236 * The checks are in order of increasing expense: checks that can be
1237 * carried out against CPU-local state are performed first. However,
1238 * we must check for CPU stalls first, else we might not get a chance.
1239 */
1240static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1241{
1242 rdp->n_rcu_pending++;
1243
1244 /* Check for CPU stalls, if enabled. */
1245 check_cpu_stall(rsp, rdp);
1246
1247 /* Is the RCU core waiting for a quiescent state from this CPU? */
1248 if (rdp->qs_pending)
1249 return 1;
1250
1251 /* Does this CPU have callbacks ready to invoke? */
1252 if (cpu_has_callbacks_ready_to_invoke(rdp))
1253 return 1;
1254
1255 /* Has RCU gone idle with this CPU needing another grace period? */
1256 if (cpu_needs_another_gp(rsp, rdp))
1257 return 1;
1258
1259 /* Has another RCU grace period completed? */
1260 if (ACCESS_ONCE(rsp->completed) != rdp->completed) /* outside of lock */
1261 return 1;
1262
1263 /* Has a new RCU grace period started? */
1264 if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) /* outside of lock */
1265 return 1;
1266
1267 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1268 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) &&
1269 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0 ||
1270 (rdp->n_rcu_pending_force_qs - rdp->n_rcu_pending) < 0))
1271 return 1;
1272
1273 /* nothing to do */
1274 return 0;
1275}
1276
1277/*
1278 * Check to see if there is any immediate RCU-related work to be done
1279 * by the current CPU, returning 1 if so. This function is part of the
1280 * RCU implementation; it is -not- an exported member of the RCU API.
1281 */
1282int rcu_pending(int cpu)
1283{
1284 return __rcu_pending(&rcu_state, &per_cpu(rcu_data, cpu)) ||
1285 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu));
1286}
1287
1288/*
1289 * Check to see if any future RCU-related work will need to be done
1290 * by the current CPU, even if none need be done immediately, returning
1291 * 1 if so. This function is part of the RCU implementation; it is -not-
1292 * an exported member of the RCU API.
1293 */
1294int rcu_needs_cpu(int cpu)
1295{
1296 /* RCU callbacks either ready or pending? */
1297 return per_cpu(rcu_data, cpu).nxtlist ||
1298 per_cpu(rcu_bh_data, cpu).nxtlist;
1299}
1300
1301/*
1302 * Initialize a CPU's per-CPU RCU data. We take this "scorched earth"
1303 * approach so that we don't have to worry about how long the CPU has
1304 * been gone, or whether it ever was online previously. We do trust the
1305 * ->mynode field, as it is constant for a given struct rcu_data and
1306 * initialized during early boot.
1307 *
1308 * Note that only one online or offline event can be happening at a given
1309 * time. Note also that we can accept some slop in the rsp->completed
1310 * access due to the fact that this CPU cannot possibly have any RCU
1311 * callbacks in flight yet.
1312 */
1313static void
1314rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
1315{
1316 unsigned long flags;
1317 int i;
1318 long lastcomp;
1319 unsigned long mask;
1320 struct rcu_data *rdp = rsp->rda[cpu];
1321 struct rcu_node *rnp = rcu_get_root(rsp);
1322
1323 /* Set up local state, ensuring consistent view of global state. */
1324 spin_lock_irqsave(&rnp->lock, flags);
1325 lastcomp = rsp->completed;
1326 rdp->completed = lastcomp;
1327 rdp->gpnum = lastcomp;
1328 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1329 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1330 rdp->beenonline = 1; /* We have now been online. */
1331 rdp->passed_quiesc_completed = lastcomp - 1;
1332 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1333 rdp->nxtlist = NULL;
1334 for (i = 0; i < RCU_NEXT_SIZE; i++)
1335 rdp->nxttail[i] = &rdp->nxtlist;
1336 rdp->qlen = 0;
1337 rdp->blimit = blimit;
1338#ifdef CONFIG_NO_HZ
1339 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1340#endif /* #ifdef CONFIG_NO_HZ */
1341 rdp->cpu = cpu;
1342 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1343
1344 /*
1345 * A new grace period might start here. If so, we won't be part
1346 * of it, but that is OK, as we are currently in a quiescent state.
1347 */
1348
1349 /* Exclude any attempts to start a new GP on large systems. */
1350 spin_lock(&rsp->onofflock); /* irqs already disabled. */
1351
1352 /* Add CPU to rcu_node bitmasks. */
1353 rnp = rdp->mynode;
1354 mask = rdp->grpmask;
1355 do {
1356 /* Exclude any attempts to start a new GP on small systems. */
1357 spin_lock(&rnp->lock); /* irqs already disabled. */
1358 rnp->qsmaskinit |= mask;
1359 mask = rnp->grpmask;
1360 spin_unlock(&rnp->lock); /* irqs already disabled. */
1361 rnp = rnp->parent;
1362 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1363
1364 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1365
1366 /*
1367 * A new grace period might start here. If so, we will be part of
1368 * it, and its gpnum will be greater than ours, so we will
1369 * participate. It is also possible for the gpnum to have been
1370 * incremented before this function was called, and the bitmasks
1371 * to not be filled out until now, in which case we will also
1372 * participate due to our gpnum being behind.
1373 */
1374
1375 /* Since it is coming online, the CPU is in a quiescent state. */
1376 cpu_quiet(cpu, rsp, rdp, lastcomp);
1377 local_irq_restore(flags);
1378}
1379
1380static void __cpuinit rcu_online_cpu(int cpu)
1381{
1382#ifdef CONFIG_NO_HZ
1383 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
1384
1385 rdtp->dynticks_nesting = 1;
1386 rdtp->dynticks |= 1; /* need consecutive #s even for hotplug. */
1387 rdtp->dynticks_nmi = (rdtp->dynticks_nmi + 1) & ~0x1;
1388#endif /* #ifdef CONFIG_NO_HZ */
1389 rcu_init_percpu_data(cpu, &rcu_state);
1390 rcu_init_percpu_data(cpu, &rcu_bh_state);
1391 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1392}
1393
1394/*
1395 * Handle CPU online/offline notifcation events.
1396 */
1397static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1398 unsigned long action, void *hcpu)
1399{
1400 long cpu = (long)hcpu;
1401
1402 switch (action) {
1403 case CPU_UP_PREPARE:
1404 case CPU_UP_PREPARE_FROZEN:
1405 rcu_online_cpu(cpu);
1406 break;
1407 case CPU_DEAD:
1408 case CPU_DEAD_FROZEN:
1409 case CPU_UP_CANCELED:
1410 case CPU_UP_CANCELED_FROZEN:
1411 rcu_offline_cpu(cpu);
1412 break;
1413 default:
1414 break;
1415 }
1416 return NOTIFY_OK;
1417}
1418
1419/*
1420 * Compute the per-level fanout, either using the exact fanout specified
1421 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1422 */
1423#ifdef CONFIG_RCU_FANOUT_EXACT
1424static void __init rcu_init_levelspread(struct rcu_state *rsp)
1425{
1426 int i;
1427
1428 for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1429 rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1430}
1431#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1432static void __init rcu_init_levelspread(struct rcu_state *rsp)
1433{
1434 int ccur;
1435 int cprv;
1436 int i;
1437
1438 cprv = NR_CPUS;
1439 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1440 ccur = rsp->levelcnt[i];
1441 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1442 cprv = ccur;
1443 }
1444}
1445#endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1446
1447/*
1448 * Helper function for rcu_init() that initializes one rcu_state structure.
1449 */
1450static void __init rcu_init_one(struct rcu_state *rsp)
1451{
1452 int cpustride = 1;
1453 int i;
1454 int j;
1455 struct rcu_node *rnp;
1456
1457 /* Initialize the level-tracking arrays. */
1458
1459 for (i = 1; i < NUM_RCU_LVLS; i++)
1460 rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1461 rcu_init_levelspread(rsp);
1462
1463 /* Initialize the elements themselves, starting from the leaves. */
1464
1465 for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1466 cpustride *= rsp->levelspread[i];
1467 rnp = rsp->level[i];
1468 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1469 spin_lock_init(&rnp->lock);
1470 rnp->qsmask = 0;
1471 rnp->qsmaskinit = 0;
1472 rnp->grplo = j * cpustride;
1473 rnp->grphi = (j + 1) * cpustride - 1;
1474 if (rnp->grphi >= NR_CPUS)
1475 rnp->grphi = NR_CPUS - 1;
1476 if (i == 0) {
1477 rnp->grpnum = 0;
1478 rnp->grpmask = 0;
1479 rnp->parent = NULL;
1480 } else {
1481 rnp->grpnum = j % rsp->levelspread[i - 1];
1482 rnp->grpmask = 1UL << rnp->grpnum;
1483 rnp->parent = rsp->level[i - 1] +
1484 j / rsp->levelspread[i - 1];
1485 }
1486 rnp->level = i;
1487 }
1488 }
1489}
1490
1491/*
1492 * Helper macro for __rcu_init(). To be used nowhere else!
1493 * Assigns leaf node pointers into each CPU's rcu_data structure.
1494 */
1495#define RCU_DATA_PTR_INIT(rsp, rcu_data) \
1496do { \
1497 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1498 j = 0; \
1499 for_each_possible_cpu(i) { \
1500 if (i > rnp[j].grphi) \
1501 j++; \
1502 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1503 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1504 } \
1505} while (0)
1506
1507static struct notifier_block __cpuinitdata rcu_nb = {
1508 .notifier_call = rcu_cpu_notify,
1509};
1510
1511void __init __rcu_init(void)
1512{
1513 int i; /* All used by RCU_DATA_PTR_INIT(). */
1514 int j;
1515 struct rcu_node *rnp;
1516
1517 printk(KERN_WARNING "Experimental hierarchical RCU implementation.\n");
1518#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1519 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1520#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1521 rcu_init_one(&rcu_state);
1522 RCU_DATA_PTR_INIT(&rcu_state, rcu_data);
1523 rcu_init_one(&rcu_bh_state);
1524 RCU_DATA_PTR_INIT(&rcu_bh_state, rcu_bh_data);
1525
1526 for_each_online_cpu(i)
1527 rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, (void *)(long)i);
1528 /* Register notifier for non-boot CPUs */
1529 register_cpu_notifier(&rcu_nb);
1530 printk(KERN_WARNING "Experimental hierarchical RCU init done.\n");
1531}
1532
1533module_param(blimit, int, 0);
1534module_param(qhimark, int, 0);
1535module_param(qlowmark, int, 0);
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-29 20:04:52 -0500