RAIDZ: Use cache blocking during parity math

RAIDZ parity is calculated by adding data one column at a time.  It
works OK for small blocks, but for large blocks results of previous
addition may already be evicted from CPU caches to main memory, and
in addition to extra memory write require extra read to get it back.

This patch splits large parity operations into 64KB chunks, that
should in most cases fit into CPU L2 caches from the last decade.
I haven't touched more complicated cases of data reconstruction to
not overcomplicate the code.  Those should be relatively rare.

My tests on Xeon Gold 6242R CPU with 1MB of L2 cache per core show
up to 10/20% memory traffic reduction when writing to 4-wide RAIDZ/
RAIDZ2 blocks of ~4MB and up.  Older CPUs with 256KB of L2 cache
should see the effect even on smaller blocks.  Wider vdevs may need
bigger blocks to be affected.

Signed-off-by:	Alexander Motin <[email protected]>
Sponsored by:	iXsystems, Inc.

Author: amotin

include/sys/abd.h

@@ -133,11 +133,11 @@ int abd_cmp_buf_off(abd_t *, const void *, size_t, size_t);
 void abd_zero_off(abd_t *, size_t, size_t);
 void abd_verify(abd_t *);
 
-void abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
-	ssize_t csize, ssize_t dsize, const unsigned parity,
+void abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd, size_t off,
+	size_t csize, size_t dsize, const unsigned parity,
 	void (*func_raidz_gen)(void **, const void *, size_t, size_t));
 void abd_raidz_rec_iterate(abd_t **cabds, abd_t **tabds,
-	ssize_t tsize, const unsigned parity,
+	size_t tsize, const unsigned parity,
 	void (*func_raidz_rec)(void **t, const size_t tsize, void **c,
 	const unsigned *mul),
 	const unsigned *mul);

module/zfs/abd.c

@@ -1017,12 +1017,12 @@ abd_cmp(abd_t *dabd, abd_t *sabd)
  *                 is the same when taking linear and when taking scatter
  */
 void
-abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
-    ssize_t csize, ssize_t dsize, const unsigned parity,
+abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd, size_t off,
+    size_t csize, size_t dsize, const unsigned parity,
     void (*func_raidz_gen)(void **, const void *, size_t, size_t))
 {
 	int i;
-	ssize_t len, dlen;
+	size_t len, dlen;
 	struct abd_iter caiters[3];
 	struct abd_iter daiter;
 	void *caddrs[3];
@@ -1033,16 +1033,15 @@ abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
 	ASSERT3U(parity, <=, 3);
 	for (i = 0; i < parity; i++) {
 		abd_verify(cabds[i]);
-		ASSERT3U(csize, <=, cabds[i]->abd_size);
-		c_cabds[i] = abd_init_abd_iter(cabds[i], &caiters[i], 0);
+		ASSERT3U(off + csize, <=, cabds[i]->abd_size);
+		c_cabds[i] = abd_init_abd_iter(cabds[i], &caiters[i], off);
 	}
 
-	ASSERT3S(dsize, >=, 0);
 	if (dsize > 0) {
 		ASSERT(dabd);
 		abd_verify(dabd);
-		ASSERT3U(dsize, <=, dabd->abd_size);
-		c_dabd = abd_init_abd_iter(dabd, &daiter, 0);
+		ASSERT3U(off + dsize, <=, dabd->abd_size);
+		c_dabd = abd_init_abd_iter(dabd, &daiter, off);
 	}
 
 	abd_enter_critical(flags);
@@ -1064,7 +1063,7 @@ abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
 			dlen = 0;
 
 		/* must be progressive */
-		ASSERT3S(len, >, 0);
+		ASSERT3U(len, >, 0);
 		/*
 		 * The iterated function likely will not do well if each
 		 * segment except the last one is not multiple of 512 (raidz).
@@ -1089,9 +1088,6 @@ abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
 		}
 
 		csize -= len;
-
-		ASSERT3S(dsize, >=, 0);
-		ASSERT3S(csize, >=, 0);
 	}
 	abd_exit_critical(flags);
 }
@@ -1108,13 +1104,13 @@ abd_raidz_gen_iterate(abd_t **cabds, abd_t *dabd,
  */
 void
 abd_raidz_rec_iterate(abd_t **cabds, abd_t **tabds,
-    ssize_t tsize, const unsigned parity,
+    size_t tsize, const unsigned parity,
     void (*func_raidz_rec)(void **t, const size_t tsize, void **c,
     const unsigned *mul),
     const unsigned *mul)
 {
 	int i;
-	ssize_t len;
+	size_t len;
 	struct abd_iter citers[3];
 	struct abd_iter xiters[3];
 	void *caddrs[3], *xaddrs[3];