spinlock.c 8.2 KB
Newer Older
1 2 3 4 5 6
/*
 * Split spinlock implementation out into its own file, so it can be
 * compiled in a FTRACE-compatible way.
 */
#include <linux/kernel_stat.h>
#include <linux/spinlock.h>
J
Jeremy Fitzhardinge 已提交
7 8
#include <linux/debugfs.h>
#include <linux/log2.h>
9
#include <linux/gfp.h>
10
#include <linux/slab.h>
11 12 13 14 15 16 17

#include <asm/paravirt.h>

#include <xen/interface/xen.h>
#include <xen/events.h>

#include "xen-ops.h"
J
Jeremy Fitzhardinge 已提交
18 19
#include "debugfs.h"

20 21 22 23 24 25 26 27
enum xen_contention_stat {
	TAKEN_SLOW,
	TAKEN_SLOW_PICKUP,
	TAKEN_SLOW_SPURIOUS,
	RELEASED_SLOW,
	RELEASED_SLOW_KICKED,
	NR_CONTENTION_STATS
};
J
Jeremy Fitzhardinge 已提交
28 29


30
#ifdef CONFIG_XEN_DEBUG_FS
31
#define HISTO_BUCKETS	30
32 33 34
static struct xen_spinlock_stats
{
	u32 contention_stats[NR_CONTENTION_STATS];
35 36
	u32 histo_spin_blocked[HISTO_BUCKETS+1];
	u64 time_blocked;
J
Jeremy Fitzhardinge 已提交
37 38 39 40 41 42
} spinlock_stats;

static u8 zero_stats;

static inline void check_zero(void)
{
43 44 45 46 47 48 49
	u8 ret;
	u8 old = ACCESS_ONCE(zero_stats);
	if (unlikely(old)) {
		ret = cmpxchg(&zero_stats, old, 0);
		/* This ensures only one fellow resets the stat */
		if (ret == old)
			memset(&spinlock_stats, 0, sizeof(spinlock_stats));
J
Jeremy Fitzhardinge 已提交
50 51 52
	}
}

53 54 55 56 57
static inline void add_stats(enum xen_contention_stat var, u32 val)
{
	check_zero();
	spinlock_stats.contention_stats[var] += val;
}
J
Jeremy Fitzhardinge 已提交
58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75

static inline u64 spin_time_start(void)
{
	return xen_clocksource_read();
}

static void __spin_time_accum(u64 delta, u32 *array)
{
	unsigned index = ilog2(delta);

	check_zero();

	if (index < HISTO_BUCKETS)
		array[index]++;
	else
		array[HISTO_BUCKETS]++;
}

76
static inline void spin_time_accum_blocked(u64 start)
J
Jeremy Fitzhardinge 已提交
77 78 79
{
	u32 delta = xen_clocksource_read() - start;

80 81
	__spin_time_accum(delta, spinlock_stats.histo_spin_blocked);
	spinlock_stats.time_blocked += delta;
J
Jeremy Fitzhardinge 已提交
82 83
}
#else  /* !CONFIG_XEN_DEBUG_FS */
84 85 86
static inline void add_stats(enum xen_contention_stat var, u32 val)
{
}
J
Jeremy Fitzhardinge 已提交
87 88 89 90 91 92

static inline u64 spin_time_start(void)
{
	return 0;
}

93
static inline void spin_time_accum_blocked(u64 start)
J
Jeremy Fitzhardinge 已提交
94 95 96
{
}
#endif  /* CONFIG_XEN_DEBUG_FS */
97

98 99 100
struct xen_lock_waiting {
	struct arch_spinlock *lock;
	__ticket_t want;
101 102
};

103
static DEFINE_PER_CPU(int, lock_kicker_irq) = -1;
104
static DEFINE_PER_CPU(char *, irq_name);
105 106
static DEFINE_PER_CPU(struct xen_lock_waiting, lock_waiting);
static cpumask_t waiting_cpus;
107

108
static bool xen_pvspin = true;
109
__visible void xen_lock_spinning(struct arch_spinlock *lock, __ticket_t want)
110
{
C
Christoph Lameter 已提交
111
	int irq = __this_cpu_read(lock_kicker_irq);
112 113
	struct xen_lock_waiting *w = &__get_cpu_var(lock_waiting);
	int cpu = smp_processor_id();
114
	u64 start;
115
	unsigned long flags;
116 117 118

	/* If kicker interrupts not initialized yet, just spin */
	if (irq == -1)
119
		return;
120

121 122
	start = spin_time_start();

123 124 125 126 127
	/*
	 * Make sure an interrupt handler can't upset things in a
	 * partially setup state.
	 */
	local_irq_save(flags);
128 129 130 131 132 133 134 135 136 137 138 139 140 141
	/*
	 * We don't really care if we're overwriting some other
	 * (lock,want) pair, as that would mean that we're currently
	 * in an interrupt context, and the outer context had
	 * interrupts enabled.  That has already kicked the VCPU out
	 * of xen_poll_irq(), so it will just return spuriously and
	 * retry with newly setup (lock,want).
	 *
	 * The ordering protocol on this is that the "lock" pointer
	 * may only be set non-NULL if the "want" ticket is correct.
	 * If we're updating "want", we must first clear "lock".
	 */
	w->lock = NULL;
	smp_wmb();
142 143 144
	w->want = want;
	smp_wmb();
	w->lock = lock;
145

146 147 148
	/* This uses set_bit, which atomic and therefore a barrier */
	cpumask_set_cpu(cpu, &waiting_cpus);
	add_stats(TAKEN_SLOW, 1);
149

150 151
	/* clear pending */
	xen_clear_irq_pending(irq);
152

153 154
	/* Only check lock once pending cleared */
	barrier();
155

156 157 158 159
	/*
	 * Mark entry to slowpath before doing the pickup test to make
	 * sure we don't deadlock with an unlocker.
	 */
160 161
	__ticket_enter_slowpath(lock);

162 163 164 165
	/*
	 * check again make sure it didn't become free while
	 * we weren't looking
	 */
166 167 168 169
	if (ACCESS_ONCE(lock->tickets.head) == want) {
		add_stats(TAKEN_SLOW_PICKUP, 1);
		goto out;
	}
170 171 172 173 174 175 176 177 178 179

	/* Allow interrupts while blocked */
	local_irq_restore(flags);

	/*
	 * If an interrupt happens here, it will leave the wakeup irq
	 * pending, which will cause xen_poll_irq() to return
	 * immediately.
	 */

180 181 182
	/* Block until irq becomes pending (or perhaps a spurious wakeup) */
	xen_poll_irq(irq);
	add_stats(TAKEN_SLOW_SPURIOUS, !xen_test_irq_pending(irq));
183 184 185

	local_irq_save(flags);

186
	kstat_incr_irq_this_cpu(irq);
187
out:
188 189
	cpumask_clear_cpu(cpu, &waiting_cpus);
	w->lock = NULL;
190

191
	local_irq_restore(flags);
192

193
	spin_time_accum_blocked(start);
194
}
195
PV_CALLEE_SAVE_REGS_THUNK(xen_lock_spinning);
196

197
static void xen_unlock_kick(struct arch_spinlock *lock, __ticket_t next)
198 199 200
{
	int cpu;

201 202 203 204
	add_stats(RELEASED_SLOW, 1);

	for_each_cpu(cpu, &waiting_cpus) {
		const struct xen_lock_waiting *w = &per_cpu(lock_waiting, cpu);
J
Jeremy Fitzhardinge 已提交
205

206 207 208
		/* Make sure we read lock before want */
		if (ACCESS_ONCE(w->lock) == lock &&
		    ACCESS_ONCE(w->want) == next) {
209
			add_stats(RELEASED_SLOW_KICKED, 1);
210
			xen_send_IPI_one(cpu, XEN_SPIN_UNLOCK_VECTOR);
211
			break;
212 213 214 215 216 217 218 219 220 221
		}
	}
}

static irqreturn_t dummy_handler(int irq, void *dev_id)
{
	BUG();
	return IRQ_HANDLED;
}

222
void xen_init_lock_cpu(int cpu)
223 224
{
	int irq;
225
	char *name;
226

227 228 229
	if (!xen_pvspin)
		return;

230
	WARN(per_cpu(lock_kicker_irq, cpu) >= 0, "spinlock on CPU%d exists on IRQ%d!\n",
231 232
	     cpu, per_cpu(lock_kicker_irq, cpu));

233 234 235 236
	name = kasprintf(GFP_KERNEL, "spinlock%d", cpu);
	irq = bind_ipi_to_irqhandler(XEN_SPIN_UNLOCK_VECTOR,
				     cpu,
				     dummy_handler,
237
				     IRQF_PERCPU|IRQF_NOBALANCING,
238 239 240 241 242 243
				     name,
				     NULL);

	if (irq >= 0) {
		disable_irq(irq); /* make sure it's never delivered */
		per_cpu(lock_kicker_irq, cpu) = irq;
244
		per_cpu(irq_name, cpu) = name;
245 246 247 248 249
	}

	printk("cpu %d spinlock event irq %d\n", cpu, irq);
}

A
Alex Nixon 已提交
250 251
void xen_uninit_lock_cpu(int cpu)
{
252 253 254
	if (!xen_pvspin)
		return;

A
Alex Nixon 已提交
255
	unbind_from_irqhandler(per_cpu(lock_kicker_irq, cpu), NULL);
256
	per_cpu(lock_kicker_irq, cpu) = -1;
257 258
	kfree(per_cpu(irq_name, cpu));
	per_cpu(irq_name, cpu) = NULL;
A
Alex Nixon 已提交
259 260
}

261

262 263 264 265 266 267 268 269
/*
 * Our init of PV spinlocks is split in two init functions due to us
 * using paravirt patching and jump labels patching and having to do
 * all of this before SMP code is invoked.
 *
 * The paravirt patching needs to be done _before_ the alternative asm code
 * is started, otherwise we would not patch the core kernel code.
 */
270 271
void __init xen_init_spinlocks(void)
{
272

273 274 275 276
	if (!xen_pvspin) {
		printk(KERN_DEBUG "xen: PV spinlocks disabled\n");
		return;
	}
277
	printk(KERN_DEBUG "xen: PV spinlocks enabled\n");
278
	pv_lock_ops.lock_spinning = PV_CALLEE_SAVE(xen_lock_spinning);
279
	pv_lock_ops.unlock_kick = xen_unlock_kick;
280
}
J
Jeremy Fitzhardinge 已提交
281

282 283 284 285 286 287 288 289 290 291 292
/*
 * While the jump_label init code needs to happend _after_ the jump labels are
 * enabled and before SMP is started. Hence we use pre-SMP initcall level
 * init. We cannot do it in xen_init_spinlocks as that is done before
 * jump labels are activated.
 */
static __init int xen_init_spinlocks_jump(void)
{
	if (!xen_pvspin)
		return 0;

293 294 295
	if (!xen_domain())
		return 0;

296 297 298 299 300
	static_key_slow_inc(&paravirt_ticketlocks_enabled);
	return 0;
}
early_initcall(xen_init_spinlocks_jump);

301 302 303 304 305 306 307
static __init int xen_parse_nopvspin(char *arg)
{
	xen_pvspin = false;
	return 0;
}
early_param("xen_nopvspin", xen_parse_nopvspin);

J
Jeremy Fitzhardinge 已提交
308 309 310 311 312 313 314 315 316 317 318
#ifdef CONFIG_XEN_DEBUG_FS

static struct dentry *d_spin_debug;

static int __init xen_spinlock_debugfs(void)
{
	struct dentry *d_xen = xen_init_debugfs();

	if (d_xen == NULL)
		return -ENOMEM;

319 320 321
	if (!xen_pvspin)
		return 0;

J
Jeremy Fitzhardinge 已提交
322 323 324 325 326
	d_spin_debug = debugfs_create_dir("spinlocks", d_xen);

	debugfs_create_u8("zero_stats", 0644, d_spin_debug, &zero_stats);

	debugfs_create_u32("taken_slow", 0444, d_spin_debug,
327
			   &spinlock_stats.contention_stats[TAKEN_SLOW]);
J
Jeremy Fitzhardinge 已提交
328
	debugfs_create_u32("taken_slow_pickup", 0444, d_spin_debug,
329
			   &spinlock_stats.contention_stats[TAKEN_SLOW_PICKUP]);
J
Jeremy Fitzhardinge 已提交
330
	debugfs_create_u32("taken_slow_spurious", 0444, d_spin_debug,
331
			   &spinlock_stats.contention_stats[TAKEN_SLOW_SPURIOUS]);
J
Jeremy Fitzhardinge 已提交
332 333

	debugfs_create_u32("released_slow", 0444, d_spin_debug,
334
			   &spinlock_stats.contention_stats[RELEASED_SLOW]);
J
Jeremy Fitzhardinge 已提交
335
	debugfs_create_u32("released_slow_kicked", 0444, d_spin_debug,
336
			   &spinlock_stats.contention_stats[RELEASED_SLOW_KICKED]);
J
Jeremy Fitzhardinge 已提交
337

338 339
	debugfs_create_u64("time_blocked", 0444, d_spin_debug,
			   &spinlock_stats.time_blocked);
J
Jeremy Fitzhardinge 已提交
340

341 342
	debugfs_create_u32_array("histo_blocked", 0444, d_spin_debug,
				spinlock_stats.histo_spin_blocked, HISTO_BUCKETS + 1);
J
Jeremy Fitzhardinge 已提交
343 344 345 346 347 348

	return 0;
}
fs_initcall(xen_spinlock_debugfs);

#endif	/* CONFIG_XEN_DEBUG_FS */