Illumos #3582, #3584

3582 zfs_delay() should support a variable resolution
3584 DTrace sdt probes for ZFS txg states

Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Dan McDonald <danmcd@nexenta.com>
Reviewed by: Richard Elling <richard.elling@dey-sys.com>
Approved by: Garrett D'Amore <garrett@damore.org>

References:
    https://www.illumos.org/issues/3582
    illumos/illumos-gate@0689f76

Ported by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #1775

include/sys/txg.h

@@ -74,13 +74,8 @@ extern void txg_rele_to_quiesce(txg_handle_t *txghp);
 extern void txg_rele_to_sync(txg_handle_t *txghp);
 extern void txg_register_callbacks(txg_handle_t *txghp, list_t *tx_callbacks);
 
-/*
+extern void txg_delay(struct dsl_pool *dp, uint64_t txg, hrtime_t delta,
- * Delay the caller by the specified number of ticks or until
+    hrtime_t resolution);
- * the txg closes (whichever comes first).  This is intended
- * to be used to throttle writers when the system nears its
- * capacity.
- */
-extern void txg_delay(struct dsl_pool *dp, uint64_t txg, int ticks);
 
 /*
  * Wait until the given transaction group has finished syncing.

include/sys/txg_impl.h

@@ -70,7 +70,7 @@ struct tx_cpu {
 	kmutex_t	tc_open_lock;	/* protects tx_open_txg */
 	kmutex_t	tc_lock;	/* protects the rest of this struct */
 	kcondvar_t	tc_cv[TXG_SIZE];
-	uint64_t	tc_count[TXG_SIZE];
+	uint64_t	tc_count[TXG_SIZE];	/* tx hold count on each txg */
 	list_t		tc_callbacks[TXG_SIZE]; /* commit cb list */
 	char		tc_pad[8];		/* pad to fill 3 cache lines */
 };
@@ -87,8 +87,8 @@ struct tx_cpu {
  * every cpu (see txg_quiesce()).
  */
 typedef struct tx_state {
-	tx_cpu_t	*tx_cpu;	/* protects right to enter txg	*/
+	tx_cpu_t	*tx_cpu;	/* protects access to tx_open_txg */
-	kmutex_t	tx_sync_lock;	/* protects tx_state_t */
+	kmutex_t	tx_sync_lock;	/* protects the rest of this struct */
 	uint64_t	tx_open_txg;	/* currently open txg id */
 	uint64_t	tx_quiesced_txg; /* quiesced txg waiting for sync */
 	uint64_t	tx_syncing_txg;	/* currently syncing txg id */

include/sys/zfs_context.h

@@ -338,6 +338,8 @@ extern void cv_init(kcondvar_t *cv, char *name, int type, void *arg);
 extern void cv_destroy(kcondvar_t *cv);
 extern void cv_wait(kcondvar_t *cv, kmutex_t *mp);
 extern clock_t cv_timedwait(kcondvar_t *cv, kmutex_t *mp, clock_t abstime);
+extern clock_t cv_timedwait_hires(kcondvar_t *cvp, kmutex_t *mp, hrtime_t tim,
+    hrtime_t res, int flag);
 extern void cv_signal(kcondvar_t *cv);
 extern void cv_broadcast(kcondvar_t *cv);
 #define	cv_timedwait_interruptible(cv, mp, at)	cv_timedwait(cv, mp, at)

lib/libspl/include/sys/time.h

@@ -50,6 +50,14 @@
 #define NSEC_PER_USEC	1000L
 #endif
 
+#ifndef MSEC2NSEC
+#define MSEC2NSEC(m)    ((hrtime_t)(m) * (NANOSEC / MILLISEC))
+#endif
+
+#ifndef NSEC2MSEC
+#define NSEC2MSEC(n)    ((n) / (NANOSEC / MILLISEC))
+#endif
+
 extern hrtime_t gethrtime(void);
 extern void gethrestime(timestruc_t *);
 

lib/libzpool/kernel.c

@@ -528,6 +528,41 @@ top:
 	return (1);
 }
 
+/*ARGSUSED*/
+clock_t
+cv_timedwait_hires(kcondvar_t *cv, kmutex_t *mp, hrtime_t tim, hrtime_t res,
+    int flag)
+{
+	int error;
+	timestruc_t ts;
+	hrtime_t delta;
+
+	ASSERT(flag == 0);
+
+top:
+	delta = tim - gethrtime();
+	if (delta <= 0)
+		return (-1);
+
+	ts.tv_sec = delta / NANOSEC;
+	ts.tv_nsec = delta % NANOSEC;
+
+	ASSERT(mutex_owner(mp) == curthread);
+	mp->m_owner = NULL;
+	error = pthread_cond_timedwait(&cv->cv, &mp->m_lock, &ts);
+	mp->m_owner = curthread;
+
+	if (error == ETIME)
+		return (-1);
+
+	if (error == EINTR)
+		goto top;
+
+	ASSERT(error == 0);
+
+	return (1);
+}
+
 void
 cv_signal(kcondvar_t *cv)
 {

module/zfs/dsl_dir.c

@@ -743,7 +743,8 @@ dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize,
 		err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx);
 	} else {
 		if (err == EAGAIN) {
-			txg_delay(dd->dd_pool, tx->tx_txg, 1);
+			txg_delay(dd->dd_pool, tx->tx_txg,
+			    MSEC2NSEC(10), MSEC2NSEC(10));
 			err = SET_ERROR(ERESTART);
 		}
 		dsl_pool_memory_pressure(dd->dd_pool);

module/zfs/dsl_pool.c

@@ -58,6 +58,9 @@ kmutex_t zfs_write_limit_lock;
 
 static pgcnt_t old_physmem = 0;
 
+hrtime_t zfs_throttle_delay = MSEC2NSEC(10);
+hrtime_t zfs_throttle_resolution = MSEC2NSEC(10);
+
 int
 dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
 {
@@ -512,12 +515,13 @@ dsl_pool_sync(dsl_pool_t *dp, uint64_t txg)
 	 * Weight the throughput calculation towards the current value:
 	 * 	thru = 3/4 old_thru + 1/4 new_thru
 	 *
-	 * Note: write_time is in nanosecs, so write_time/MICROSEC
+	 * Note: write_time is in nanosecs while dp_throughput is expressed in
-	 * yields millisecs
+	 * bytes per millisecond.
 	 */
 	ASSERT(zfs_write_limit_min > 0);
-	if (data_written > zfs_write_limit_min / 8 && write_time > MICROSEC) {
+	if (data_written > zfs_write_limit_min / 8 &&
-		uint64_t throughput = data_written / (write_time / MICROSEC);
+	    write_time > MSEC2NSEC(1)) {
+		uint64_t throughput = data_written / NSEC2MSEC(write_time);
 
 		if (dp->dp_throughput)
 			dp->dp_throughput = throughput / 4 +
@@ -617,8 +621,10 @@ dsl_pool_tempreserve_space(dsl_pool_t *dp, uint64_t space, dmu_tx_t *tx)
 	 * the caller 1 clock tick.  This will slow down the "fill"
 	 * rate until the sync process can catch up with us.
 	 */
-	if (reserved && reserved > (write_limit - (write_limit >> 3)))
+	if (reserved && reserved > (write_limit - (write_limit >> 3))) {
-		txg_delay(dp, tx->tx_txg, 1);
+		txg_delay(dp, tx->tx_txg, zfs_throttle_delay,
+		    zfs_throttle_resolution);
+	}
 
 	return (0);
 }

module/zfs/dsl_scan.c

@@ -409,7 +409,7 @@ dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_t *zb)
 	    zfs_resilver_min_time_ms : zfs_scan_min_time_ms;
 	elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
 	if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
-	    (elapsed_nanosecs / MICROSEC > mintime &&
+	    (NSEC2MSEC(elapsed_nanosecs) > mintime &&
 	    txg_sync_waiting(scn->scn_dp)) ||
 	    spa_shutting_down(scn->scn_dp->dp_spa)) {
 		if (zb) {
@@ -1335,7 +1335,7 @@ dsl_scan_free_should_pause(dsl_scan_t *scn)
 
 	elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
 	return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
-	    (elapsed_nanosecs / MICROSEC > zfs_free_min_time_ms &&
+	    (NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms &&
 	    txg_sync_waiting(scn->scn_dp)) ||
 	    spa_shutting_down(scn->scn_dp->dp_spa));
 }
@@ -1459,7 +1459,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 			    "free_bpobj/bptree txg %llu",
 			    (longlong_t)scn->scn_visited_this_txg,
 			    (longlong_t)
-			    (gethrtime() - scn->scn_sync_start_time) / MICROSEC,
+			    NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
 			    (longlong_t)tx->tx_txg);
 			scn->scn_visited_this_txg = 0;
 			/*
@@ -1507,7 +1507,7 @@ dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
 
 	zfs_dbgmsg("visited %llu blocks in %llums",
 	    (longlong_t)scn->scn_visited_this_txg,
-	    (longlong_t)(gethrtime() - scn->scn_sync_start_time) / MICROSEC);
+	    (longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time));
 
 	if (!scn->scn_pausing) {
 		/* finished with scan. */

module/zfs/spa_misc.c

@@ -490,8 +490,8 @@ spa_add(const char *name, nvlist_t *config, const char *altroot)
 	spa->spa_proc = &p0;
 	spa->spa_proc_state = SPA_PROC_NONE;
 
-	spa->spa_deadman_synctime = zfs_deadman_synctime *
+	spa->spa_deadman_synctime = MSEC2NSEC(zfs_deadman_synctime *
-	    zfs_txg_synctime_ms * MICROSEC;
+	    zfs_txg_synctime_ms);
 
 	refcount_create(&spa->spa_refcount);
 	spa_config_lock_init(spa);

module/zfs/txg.c

@@ -236,7 +236,7 @@ txg_thread_exit(tx_state_t *tx, callb_cpr_t *cpr, kthread_t **tpp)
 }
 
 static void
-txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, uint64_t time)
+txg_thread_wait(tx_state_t *tx, callb_cpr_t *cpr, kcondvar_t *cv, clock_t time)
 {
 	CALLB_CPR_SAFE_BEGIN(cpr);
 
@@ -373,6 +373,9 @@ txg_quiesce(dsl_pool_t *dp, uint64_t txg)
 	spa_txg_history_set(dp->dp_spa, txg, TXG_STATE_OPEN, gethrtime());
 	spa_txg_history_add(dp->dp_spa, tx->tx_open_txg);
 
+	DTRACE_PROBE2(txg__quiescing, dsl_pool_t *, dp, uint64_t, txg);
+	DTRACE_PROBE2(txg__opened, dsl_pool_t *, dp, uint64_t, tx->tx_open_txg);
+
 	/*
 	 * Now that we've incremented tx_open_txg, we can let threads
 	 * enter the next transaction group.
@@ -531,6 +534,7 @@ txg_sync_thread(dsl_pool_t *dp)
 		txg = tx->tx_quiesced_txg;
 		tx->tx_quiesced_txg = 0;
 		tx->tx_syncing_txg = txg;
+		DTRACE_PROBE2(txg__syncing, dsl_pool_t *, dp, uint64_t, txg);
 		cv_broadcast(&tx->tx_quiesce_more_cv);
 
 		dprintf("txg=%llu quiesce_txg=%llu sync_txg=%llu\n",
@@ -544,6 +548,7 @@ txg_sync_thread(dsl_pool_t *dp)
 		mutex_enter(&tx->tx_sync_lock);
 		tx->tx_synced_txg = txg;
 		tx->tx_syncing_txg = 0;
+		DTRACE_PROBE2(txg__synced, dsl_pool_t *, dp, uint64_t, txg);
 		cv_broadcast(&tx->tx_sync_done_cv);
 
 		/*
@@ -602,21 +607,22 @@ txg_quiesce_thread(dsl_pool_t *dp)
 		 */
 		dprintf("quiesce done, handing off txg %llu\n", txg);
 		tx->tx_quiesced_txg = txg;
+		DTRACE_PROBE2(txg__quiesced, dsl_pool_t *, dp, uint64_t, txg);
 		cv_broadcast(&tx->tx_sync_more_cv);
 		cv_broadcast(&tx->tx_quiesce_done_cv);
 	}
 }
 
 /*
- * Delay this thread by 'ticks' if we are still in the open transaction
+ * Delay this thread by delay nanoseconds if we are still in the open
- * group and there is already a waiting txg quiesing or quiesced.  Abort
+ * transaction group and there is already a waiting txg quiesing or quiesced.
- * the delay if this txg stalls or enters the quiesing state.
+ * Abort the delay if this txg stalls or enters the quiesing state.
  */
 void
-txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks)
+txg_delay(dsl_pool_t *dp, uint64_t txg, hrtime_t delay, hrtime_t resolution)
 {
 	tx_state_t *tx = &dp->dp_tx;
-	clock_t timeout = ddi_get_lbolt() + ticks;
+	hrtime_t start = gethrtime();
 
 	/* don't delay if this txg could transition to quiesing immediately */
 	if (tx->tx_open_txg > txg ||
@@ -629,10 +635,11 @@ txg_delay(dsl_pool_t *dp, uint64_t txg, int ticks)
 		return;
 	}
 
-	while (ddi_get_lbolt() < timeout &&
+	while (gethrtime() - start < delay &&
-	    tx->tx_syncing_txg < txg-1 && !txg_stalled(dp))
+	    tx->tx_syncing_txg < txg-1 && !txg_stalled(dp)) {
-		(void) cv_timedwait(&tx->tx_quiesce_more_cv, &tx->tx_sync_lock,
+		(void) cv_timedwait_hires(&tx->tx_quiesce_more_cv,
-		    timeout);
+		    &tx->tx_sync_lock, delay, resolution, 0);
+	}
 
 	DMU_TX_STAT_BUMP(dmu_tx_delay);