ioat(4): Add ioatcontrol(8) testing for copy_8k

Add -E ("Eight k") and -m ("Memcpy") modes to the ioatcontrol(8) tool.

Prompted by:	rpokala
Sponsored by:	EMC / Isilon Storage Division

4 files changed, 77 insertions, 15 deletions

diff --git a/sys/dev/ioat/ioat_test.c b/sys/dev/ioat/ioat_test.c
index a8311ab..f39786e 100644
--- a/sys/dev/ioat/ioat_test.c
+++ b/sys/dev/ioat/ioat_test.c
@@ -84,11 +84,17 @@ static inline void _ioat_test_log(int verbosity, const char *fmt, ...);
 static void
 ioat_test_transaction_destroy(struct test_transaction *tx)
 {
+	struct ioat_test *test;
 	int i;
 
+	test = tx->test;
+
 	for (i = 0; i < IOAT_MAX_BUFS; i++) {
 		if (tx->buf[i] != NULL) {
-			contigfree(tx->buf[i], tx->length, M_IOAT_TEST);
+			if (test->testkind == IOAT_TEST_DMA_8K)
+				free(tx->buf[i], M_IOAT_TEST);
+			else
+				contigfree(tx->buf[i], tx->length, M_IOAT_TEST);
 			tx->buf[i] = NULL;
 		}
 	}
@@ -97,8 +103,8 @@ ioat_test_transaction_destroy(struct test_transaction *tx)
 }
 
 static struct
-test_transaction *ioat_test_transaction_create(unsigned num_buffers,
-    uint32_t buffer_size)
+test_transaction *ioat_test_transaction_create(struct ioat_test *test,
+    unsigned num_buffers)
 {
 	struct test_transaction *tx;
 	unsigned i;
@@ -107,11 +113,16 @@ test_transaction *ioat_test_transaction_create(unsigned num_buffers,
 	if (tx == NULL)
 		return (NULL);
 
-	tx->length = buffer_size;
+	tx->length = test->buffer_size;
 
 	for (i = 0; i < num_buffers; i++) {
-		tx->buf[i] = contigmalloc(buffer_size, M_IOAT_TEST, M_NOWAIT,
-		    0, BUS_SPACE_MAXADDR, PAGE_SIZE, 0);
+		if (test->testkind == IOAT_TEST_DMA_8K)
+			tx->buf[i] = malloc(test->buffer_size, M_IOAT_TEST,
+			    M_NOWAIT);
+		else
+			tx->buf[i] = contigmalloc(test->buffer_size,
+			    M_IOAT_TEST, M_NOWAIT, 0, BUS_SPACE_MAXADDR,
+			    PAGE_SIZE, 0);
 
 		if (tx->buf[i] == NULL) {
 			ioat_test_transaction_destroy(tx);
@@ -197,8 +208,7 @@ ioat_test_prealloc_memory(struct ioat_test *test, int index)
 	struct test_transaction *tx;
 
 	for (i = 0; i < test->transactions; i++) {
-		tx = ioat_test_transaction_create(test->chain_depth * 2,
-		    test->buffer_size);
+		tx = ioat_test_transaction_create(test, test->chain_depth * 2);
 		if (tx == NULL) {
 			ioat_test_log(0, "tx == NULL - memory exhausted\n");
 			test->status[IOAT_TEST_NO_MEMORY]++;
@@ -258,8 +268,16 @@ ioat_test_submit_1_tx(struct ioat_test *test, bus_dmaengine_t dma)
 	TAILQ_INSERT_HEAD(&test->pend_q, tx, entry);
 	IT_UNLOCK();
 
-	ioat_acquire(dma);
+	if (test->testkind != IOAT_TEST_MEMCPY)
+		ioat_acquire(dma);
 	for (i = 0; i < tx->depth; i++) {
+		if (test->testkind == IOAT_TEST_MEMCPY) {
+			memcpy(tx->buf[2 * i + 1], tx->buf[2 * i], tx->length);
+			if (i == tx->depth - 1)
+				ioat_dma_test_callback(tx, 0);
+			continue;
+		}
+
 		src = vtophys((vm_offset_t)tx->buf[2*i]);
 		dest = vtophys((vm_offset_t)tx->buf[2*i+1]);
 
@@ -286,10 +304,20 @@ ioat_test_submit_1_tx(struct ioat_test *test, bus_dmaengine_t dma)
 			fillpattern = *(uint64_t *)tx->buf[2*i];
 			desc = ioat_blockfill(dma, dest, fillpattern,
 			    tx->length, cb, tx, flags);
+		} else if (test->testkind == IOAT_TEST_DMA_8K) {
+			bus_addr_t src2, dst2;
+
+			src2 = vtophys((vm_offset_t)tx->buf[2*i] + PAGE_SIZE);
+			dst2 = vtophys((vm_offset_t)tx->buf[2*i+1] + PAGE_SIZE);
+
+			desc = ioat_copy_8k_aligned(dma, dest, dst2, src, src2,
+			    cb, tx, flags);
 		}
 		if (desc == NULL)
 			break;
 	}
+	if (test->testkind == IOAT_TEST_MEMCPY)
+		return;
 	ioat_release(dma);
 
 	/*
@@ -317,6 +345,13 @@ ioat_dma_test(void *arg)
 	test = arg;
 	memset(__DEVOLATILE(void *, test->status), 0, sizeof(test->status));
 
+	if (test->testkind == IOAT_TEST_DMA_8K &&
+	    test->buffer_size != 2 * PAGE_SIZE) {
+		ioat_test_log(0, "Asked for 8k test and buffer size isn't 8k\n");
+		test->status[IOAT_TEST_INVALID_INPUT]++;
+		return;
+	}
+
 	if (test->buffer_size > 1024 * 1024) {
 		ioat_test_log(0, "Buffer size too large >1MB\n");
 		test->status[IOAT_TEST_NO_MEMORY]++;
diff --git a/sys/dev/ioat/ioat_test.h b/sys/dev/ioat/ioat_test.h
index b6e6a15..4306b3b 100644
--- a/sys/dev/ioat/ioat_test.h
+++ b/sys/dev/ioat/ioat_test.h
@@ -42,6 +42,8 @@ enum ioat_test_kind {
 	IOAT_TEST_FILL = 0,
 	IOAT_TEST_DMA,
 	IOAT_TEST_RAW_DMA,
+	IOAT_TEST_DMA_8K,
+	IOAT_TEST_MEMCPY,
 	IOAT_NUM_TESTKINDS
 };
 
diff --git a/tools/tools/ioat/ioatcontrol.8 b/tools/tools/ioat/ioatcontrol.8
index 2306d38..002759f 100644
--- a/tools/tools/ioat/ioatcontrol.8
+++ b/tools/tools/ioat/ioatcontrol.8
@@ -24,7 +24,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd October 28, 2015
+.Dd December 9, 2015
 .Dt IOATCONTROL 8
 .Os
 .Sh NAME
@@ -33,7 +33,9 @@
 .Xr ioat 4
 .Sh SYNOPSIS
 .Nm
+.Op Fl E
 .Op Fl f
+.Op Fl m
 .Op Fl V
 .Ar channel_number
 .Ar num_txns
@@ -55,10 +57,14 @@ allows one to issue some number of test operations to the
 driver on a specific hardware channel.
 The arguments are as follows:
 .Bl -tag -width Ds
+.It Fl E
+Test non-contiguous 8k copy.
 .It Fl f
 Test block fill (by default,
 .Nm
 tests copy)
+.It Fl m
+Test memcpy instead of DMA.
 .It Fl V
 Verify copies/fills for accuracy
 .El
diff --git a/tools/tools/ioat/ioatcontrol.c b/tools/tools/ioat/ioatcontrol.c
index 90255e7..d40ae15 100644
--- a/tools/tools/ioat/ioatcontrol.c
+++ b/tools/tools/ioat/ioatcontrol.c
@@ -48,7 +48,7 @@ static void
 usage(void)
 {
 
-	printf("Usage: %s [-fV] <channel #> <txns> [<bufsize> "
+	printf("Usage: %s [-E|-f|-m] [-V] <channel #> <txns> [<bufsize> "
 	    "[<chain-len> [duration]]]\n", getprogname());
 	printf("       %s -r [-vV] <channel #> <addr> [<bufsize>]\n",
 	    getprogname());
@@ -97,15 +97,29 @@ main(int argc, char **argv)
 {
 	struct ioat_test t;
 	int fd, ch;
-	bool fflag, rflag;
+	bool fflag, rflag, Eflag, mflag;
+	unsigned modeflags;
 
-	while ((ch = getopt(argc, argv, "rfvVw")) != -1) {
+	fflag = rflag = Eflag = mflag = false;
+	modeflags = 0;
+
+	while ((ch = getopt(argc, argv, "EfmrvVw")) != -1) {
 		switch (ch) {
+		case 'E':
+			Eflag = true;
+			modeflags++;
+			break;
 		case 'f':
 			fflag = true;
+			modeflags++;
+			break;
+		case 'm':
+			mflag = true;
+			modeflags++;
 			break;
 		case 'r':
 			rflag = true;
+			modeflags++;
 			break;
 		case 'v':
 			t.raw_is_virtual = true;
@@ -126,8 +140,8 @@ main(int argc, char **argv)
 	if (argc < 2)
 		usage();
 
-	if (rflag && fflag) {
-		printf("Invalid: -r and -f\n");
+	if (modeflags > 1) {
+		printf("Invalid: Cannot use >1 mode flag (-E, -f, -m, or -r)\n");
 		usage();
 	}
 
@@ -139,6 +153,11 @@ main(int argc, char **argv)
 
 	if (fflag)
 		t.testkind = IOAT_TEST_FILL;
+	else if (Eflag) {
+		t.testkind = IOAT_TEST_DMA_8K;
+		t.buffer_size = 8 * 1024;
+	} else if (mflag)
+		t.testkind = IOAT_TEST_MEMCPY;
 
 	t.channel_index = atoi(argv[0]);
 	if (t.channel_index > 8) {