[34] | 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 (C) IBM Corporation, 2001 |
---|
| 19 | * |
---|
| 20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> |
---|
| 21 | * Manfred Spraul <manfred@colorfullife.com> |
---|
| 22 | * |
---|
| 23 | * Modifications for Xen: Jose Renato Santos |
---|
| 24 | * Copyright (C) Hewlett-Packard, 2006 |
---|
| 25 | * |
---|
| 26 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> |
---|
| 27 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. |
---|
| 28 | * Papers: |
---|
| 29 | * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf |
---|
| 30 | * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) |
---|
| 31 | * |
---|
| 32 | * For detailed explanation of Read-Copy Update mechanism see - |
---|
| 33 | * http://lse.sourceforge.net/locking/rcupdate.html |
---|
| 34 | */ |
---|
| 35 | #include <xen/types.h> |
---|
| 36 | #include <xen/kernel.h> |
---|
| 37 | #include <xen/init.h> |
---|
| 38 | #include <xen/spinlock.h> |
---|
| 39 | #include <xen/smp.h> |
---|
| 40 | #include <xen/rcupdate.h> |
---|
| 41 | #include <xen/sched.h> |
---|
| 42 | #include <asm/atomic.h> |
---|
| 43 | #include <xen/bitops.h> |
---|
| 44 | #include <xen/percpu.h> |
---|
| 45 | #include <xen/softirq.h> |
---|
| 46 | |
---|
| 47 | /* Definition for rcupdate control block. */ |
---|
| 48 | struct rcu_ctrlblk rcu_ctrlblk = { |
---|
| 49 | .cur = -300, |
---|
| 50 | .completed = -300, |
---|
| 51 | .lock = SPIN_LOCK_UNLOCKED, |
---|
| 52 | .cpumask = CPU_MASK_NONE, |
---|
| 53 | }; |
---|
| 54 | |
---|
| 55 | DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L }; |
---|
| 56 | |
---|
| 57 | static int blimit = 10; |
---|
| 58 | static int qhimark = 10000; |
---|
| 59 | static int qlowmark = 100; |
---|
| 60 | static int rsinterval = 1000; |
---|
| 61 | |
---|
| 62 | static void force_quiescent_state(struct rcu_data *rdp, |
---|
| 63 | struct rcu_ctrlblk *rcp) |
---|
| 64 | { |
---|
| 65 | cpumask_t cpumask; |
---|
| 66 | raise_softirq(SCHEDULE_SOFTIRQ); |
---|
| 67 | if (unlikely(rdp->qlen - rdp->last_rs_qlen > rsinterval)) { |
---|
| 68 | rdp->last_rs_qlen = rdp->qlen; |
---|
| 69 | /* |
---|
| 70 | * Don't send IPI to itself. With irqs disabled, |
---|
| 71 | * rdp->cpu is the current cpu. |
---|
| 72 | */ |
---|
| 73 | cpumask = rcp->cpumask; |
---|
| 74 | cpu_clear(rdp->cpu, cpumask); |
---|
| 75 | cpumask_raise_softirq(cpumask, SCHEDULE_SOFTIRQ); |
---|
| 76 | } |
---|
| 77 | } |
---|
| 78 | |
---|
| 79 | /** |
---|
| 80 | * call_rcu - Queue an RCU callback for invocation after a grace period. |
---|
| 81 | * @head: structure to be used for queueing the RCU updates. |
---|
| 82 | * @func: actual update function to be invoked after the grace period |
---|
| 83 | * |
---|
| 84 | * The update function will be invoked some time after a full grace |
---|
| 85 | * period elapses, in other words after all currently executing RCU |
---|
| 86 | * read-side critical sections have completed. RCU read-side critical |
---|
| 87 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), |
---|
| 88 | * and may be nested. |
---|
| 89 | */ |
---|
| 90 | void fastcall call_rcu(struct rcu_head *head, |
---|
| 91 | void (*func)(struct rcu_head *rcu)) |
---|
| 92 | { |
---|
| 93 | unsigned long flags; |
---|
| 94 | struct rcu_data *rdp; |
---|
| 95 | |
---|
| 96 | head->func = func; |
---|
| 97 | head->next = NULL; |
---|
| 98 | local_irq_save(flags); |
---|
| 99 | rdp = &__get_cpu_var(rcu_data); |
---|
| 100 | *rdp->nxttail = head; |
---|
| 101 | rdp->nxttail = &head->next; |
---|
| 102 | if (unlikely(++rdp->qlen > qhimark)) { |
---|
| 103 | rdp->blimit = INT_MAX; |
---|
| 104 | force_quiescent_state(rdp, &rcu_ctrlblk); |
---|
| 105 | } |
---|
| 106 | local_irq_restore(flags); |
---|
| 107 | } |
---|
| 108 | |
---|
| 109 | /* |
---|
| 110 | * Invoke the completed RCU callbacks. They are expected to be in |
---|
| 111 | * a per-cpu list. |
---|
| 112 | */ |
---|
| 113 | static void rcu_do_batch(struct rcu_data *rdp) |
---|
| 114 | { |
---|
| 115 | struct rcu_head *next, *list; |
---|
| 116 | int count = 0; |
---|
| 117 | |
---|
| 118 | list = rdp->donelist; |
---|
| 119 | while (list) { |
---|
| 120 | next = rdp->donelist = list->next; |
---|
| 121 | list->func(list); |
---|
| 122 | list = next; |
---|
| 123 | rdp->qlen--; |
---|
| 124 | if (++count >= rdp->blimit) |
---|
| 125 | break; |
---|
| 126 | } |
---|
| 127 | if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark) |
---|
| 128 | rdp->blimit = blimit; |
---|
| 129 | if (!rdp->donelist) |
---|
| 130 | rdp->donetail = &rdp->donelist; |
---|
| 131 | else |
---|
| 132 | raise_softirq(RCU_SOFTIRQ); |
---|
| 133 | } |
---|
| 134 | |
---|
| 135 | /* |
---|
| 136 | * Grace period handling: |
---|
| 137 | * The grace period handling consists out of two steps: |
---|
| 138 | * - A new grace period is started. |
---|
| 139 | * This is done by rcu_start_batch. The start is not broadcasted to |
---|
| 140 | * all cpus, they must pick this up by comparing rcp->cur with |
---|
| 141 | * rdp->quiescbatch. All cpus are recorded in the |
---|
| 142 | * rcu_ctrlblk.cpumask bitmap. |
---|
| 143 | * - All cpus must go through a quiescent state. |
---|
| 144 | * Since the start of the grace period is not broadcasted, at least two |
---|
| 145 | * calls to rcu_check_quiescent_state are required: |
---|
| 146 | * The first call just notices that a new grace period is running. The |
---|
| 147 | * following calls check if there was a quiescent state since the beginning |
---|
| 148 | * of the grace period. If so, it updates rcu_ctrlblk.cpumask. If |
---|
| 149 | * the bitmap is empty, then the grace period is completed. |
---|
| 150 | * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace |
---|
| 151 | * period (if necessary). |
---|
| 152 | */ |
---|
| 153 | /* |
---|
| 154 | * Register a new batch of callbacks, and start it up if there is currently no |
---|
| 155 | * active batch and the batch to be registered has not already occurred. |
---|
| 156 | * Caller must hold rcu_ctrlblk.lock. |
---|
| 157 | */ |
---|
| 158 | static void rcu_start_batch(struct rcu_ctrlblk *rcp) |
---|
| 159 | { |
---|
| 160 | if (rcp->next_pending && |
---|
| 161 | rcp->completed == rcp->cur) { |
---|
| 162 | rcp->next_pending = 0; |
---|
| 163 | /* |
---|
| 164 | * next_pending == 0 must be visible in |
---|
| 165 | * __rcu_process_callbacks() before it can see new value of cur. |
---|
| 166 | */ |
---|
| 167 | smp_wmb(); |
---|
| 168 | rcp->cur++; |
---|
| 169 | |
---|
| 170 | rcp->cpumask = cpu_online_map; |
---|
| 171 | } |
---|
| 172 | } |
---|
| 173 | |
---|
| 174 | /* |
---|
| 175 | * cpu went through a quiescent state since the beginning of the grace period. |
---|
| 176 | * Clear it from the cpu mask and complete the grace period if it was the last |
---|
| 177 | * cpu. Start another grace period if someone has further entries pending |
---|
| 178 | */ |
---|
| 179 | static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp) |
---|
| 180 | { |
---|
| 181 | cpu_clear(cpu, rcp->cpumask); |
---|
| 182 | if (cpus_empty(rcp->cpumask)) { |
---|
| 183 | /* batch completed ! */ |
---|
| 184 | rcp->completed = rcp->cur; |
---|
| 185 | rcu_start_batch(rcp); |
---|
| 186 | } |
---|
| 187 | } |
---|
| 188 | |
---|
| 189 | /* |
---|
| 190 | * Check if the cpu has gone through a quiescent state (say context |
---|
| 191 | * switch). If so and if it already hasn't done so in this RCU |
---|
| 192 | * quiescent cycle, then indicate that it has done so. |
---|
| 193 | */ |
---|
| 194 | static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, |
---|
| 195 | struct rcu_data *rdp) |
---|
| 196 | { |
---|
| 197 | if (rdp->quiescbatch != rcp->cur) { |
---|
| 198 | /* start new grace period: */ |
---|
| 199 | rdp->qs_pending = 1; |
---|
| 200 | rdp->quiescbatch = rcp->cur; |
---|
| 201 | return; |
---|
| 202 | } |
---|
| 203 | |
---|
| 204 | /* Grace period already completed for this cpu? |
---|
| 205 | * qs_pending is checked instead of the actual bitmap to avoid |
---|
| 206 | * cacheline trashing. |
---|
| 207 | */ |
---|
| 208 | if (!rdp->qs_pending) |
---|
| 209 | return; |
---|
| 210 | |
---|
| 211 | rdp->qs_pending = 0; |
---|
| 212 | |
---|
| 213 | spin_lock(&rcp->lock); |
---|
| 214 | /* |
---|
| 215 | * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync |
---|
| 216 | * during cpu startup. Ignore the quiescent state. |
---|
| 217 | */ |
---|
| 218 | if (likely(rdp->quiescbatch == rcp->cur)) |
---|
| 219 | cpu_quiet(rdp->cpu, rcp); |
---|
| 220 | |
---|
| 221 | spin_unlock(&rcp->lock); |
---|
| 222 | } |
---|
| 223 | |
---|
| 224 | |
---|
| 225 | /* |
---|
| 226 | * This does the RCU processing work from softirq context. |
---|
| 227 | */ |
---|
| 228 | static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, |
---|
| 229 | struct rcu_data *rdp) |
---|
| 230 | { |
---|
| 231 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) { |
---|
| 232 | *rdp->donetail = rdp->curlist; |
---|
| 233 | rdp->donetail = rdp->curtail; |
---|
| 234 | rdp->curlist = NULL; |
---|
| 235 | rdp->curtail = &rdp->curlist; |
---|
| 236 | } |
---|
| 237 | |
---|
| 238 | local_irq_disable(); |
---|
| 239 | if (rdp->nxtlist && !rdp->curlist) { |
---|
| 240 | rdp->curlist = rdp->nxtlist; |
---|
| 241 | rdp->curtail = rdp->nxttail; |
---|
| 242 | rdp->nxtlist = NULL; |
---|
| 243 | rdp->nxttail = &rdp->nxtlist; |
---|
| 244 | local_irq_enable(); |
---|
| 245 | |
---|
| 246 | /* |
---|
| 247 | * start the next batch of callbacks |
---|
| 248 | */ |
---|
| 249 | |
---|
| 250 | /* determine batch number */ |
---|
| 251 | rdp->batch = rcp->cur + 1; |
---|
| 252 | /* see the comment and corresponding wmb() in |
---|
| 253 | * the rcu_start_batch() |
---|
| 254 | */ |
---|
| 255 | smp_rmb(); |
---|
| 256 | |
---|
| 257 | if (!rcp->next_pending) { |
---|
| 258 | /* and start it/schedule start if it's a new batch */ |
---|
| 259 | spin_lock(&rcp->lock); |
---|
| 260 | rcp->next_pending = 1; |
---|
| 261 | rcu_start_batch(rcp); |
---|
| 262 | spin_unlock(&rcp->lock); |
---|
| 263 | } |
---|
| 264 | } else { |
---|
| 265 | local_irq_enable(); |
---|
| 266 | } |
---|
| 267 | rcu_check_quiescent_state(rcp, rdp); |
---|
| 268 | if (rdp->donelist) |
---|
| 269 | rcu_do_batch(rdp); |
---|
| 270 | } |
---|
| 271 | |
---|
| 272 | static void rcu_process_callbacks(void) |
---|
| 273 | { |
---|
| 274 | __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data)); |
---|
| 275 | } |
---|
| 276 | |
---|
| 277 | static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
---|
| 278 | { |
---|
| 279 | /* This cpu has pending rcu entries and the grace period |
---|
| 280 | * for them has completed. |
---|
| 281 | */ |
---|
| 282 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) |
---|
| 283 | return 1; |
---|
| 284 | |
---|
| 285 | /* This cpu has no pending entries, but there are new entries */ |
---|
| 286 | if (!rdp->curlist && rdp->nxtlist) |
---|
| 287 | return 1; |
---|
| 288 | |
---|
| 289 | /* This cpu has finished callbacks to invoke */ |
---|
| 290 | if (rdp->donelist) |
---|
| 291 | return 1; |
---|
| 292 | |
---|
| 293 | /* The rcu core waits for a quiescent state from the cpu */ |
---|
| 294 | if (rdp->quiescbatch != rcp->cur || rdp->qs_pending) |
---|
| 295 | return 1; |
---|
| 296 | |
---|
| 297 | /* nothing to do */ |
---|
| 298 | return 0; |
---|
| 299 | } |
---|
| 300 | |
---|
| 301 | int rcu_pending(int cpu) |
---|
| 302 | { |
---|
| 303 | return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)); |
---|
| 304 | } |
---|
| 305 | |
---|
| 306 | /* |
---|
| 307 | * Check to see if any future RCU-related work will need to be done |
---|
| 308 | * by the current CPU, even if none need be done immediately, returning |
---|
| 309 | * 1 if so. This function is part of the RCU implementation; it is -not- |
---|
| 310 | * an exported member of the RCU API. |
---|
| 311 | */ |
---|
| 312 | int rcu_needs_cpu(int cpu) |
---|
| 313 | { |
---|
| 314 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
---|
| 315 | |
---|
| 316 | return (!!rdp->curlist || rcu_pending(cpu)); |
---|
| 317 | } |
---|
| 318 | |
---|
| 319 | void rcu_check_callbacks(int cpu) |
---|
| 320 | { |
---|
| 321 | raise_softirq(RCU_SOFTIRQ); |
---|
| 322 | } |
---|
| 323 | |
---|
| 324 | static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, |
---|
| 325 | struct rcu_data *rdp) |
---|
| 326 | { |
---|
| 327 | memset(rdp, 0, sizeof(*rdp)); |
---|
| 328 | rdp->curtail = &rdp->curlist; |
---|
| 329 | rdp->nxttail = &rdp->nxtlist; |
---|
| 330 | rdp->donetail = &rdp->donelist; |
---|
| 331 | rdp->quiescbatch = rcp->completed; |
---|
| 332 | rdp->qs_pending = 0; |
---|
| 333 | rdp->cpu = cpu; |
---|
| 334 | rdp->blimit = blimit; |
---|
| 335 | } |
---|
| 336 | |
---|
| 337 | void __devinit rcu_online_cpu(int cpu) |
---|
| 338 | { |
---|
| 339 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
---|
| 340 | |
---|
| 341 | rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp); |
---|
| 342 | } |
---|
| 343 | |
---|
| 344 | void rcu_init(void) |
---|
| 345 | { |
---|
| 346 | rcu_online_cpu(smp_processor_id()); |
---|
| 347 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
---|
| 348 | } |
---|