[2/2] app/compress-perf: add external mbufs option

diff --git a/app/test-compress-perf/comp_perf_options.h b/app/test-compress-perf/comp_perf_options.h
index 651cbea4e..606268373 100644
--- a/app/test-compress-perf/comp_perf_options.h
+++ b/app/test-compress-perf/comp_perf_options.h
@@ -56,12 +56,14 @@  struct comp_test_data {
 	uint32_t pool_sz;
 	uint32_t num_iter;
 	uint16_t max_sgl_segs;
+	uint32_t total_segs;
 
 	enum rte_comp_huffman huffman_enc;
 	enum comp_operation test_op;
 	int window_sz;
 	struct range_list level_lst;
 	uint8_t level;
+	int use_external_mbufs;
 
 	double ratio;
 	enum cleanup_st cleanup;
diff --git a/app/test-compress-perf/comp_perf_options_parse.c b/app/test-compress-perf/comp_perf_options_parse.c
index 01e19eafb..e623fea7f 100644
--- a/app/test-compress-perf/comp_perf_options_parse.c
+++ b/app/test-compress-perf/comp_perf_options_parse.c
@@ -28,6 +28,7 @@ 
 #define CPERF_HUFFMAN_ENC	("huffman-enc")
 #define CPERF_LEVEL		("compress-level")
 #define CPERF_WINDOW_SIZE	("window-sz")
+#define CPERF_EXTERNAL_MBUFS	("external-mbufs")
 
 struct name_id_map {
 	const char *name;
@@ -58,6 +59,8 @@  usage(char *progname)
 		"		(default: range between 1 and 9)\n"
 		" --window-sz N: base two log value of compression window size\n"
 		"		(e.g.: 15 => 32k, default: max supported by PMD)\n"
+		" --external-mbufs: use memzones as external buffers instead of\n"
+		"		keeping the data directly in mbuf area\n"
 		" -h: prints this help\n",
 		progname);
 }
@@ -520,6 +523,14 @@  parse_level(struct comp_test_data *test_data, const char *arg)
 	return 0;
 }
 
+static int
+parse_external_mbufs(struct comp_test_data *test_data,
+		     const char *arg __rte_unused)
+{
+	test_data->use_external_mbufs = 1;
+	return 0;
+}
+
 typedef int (*option_parser_t)(struct comp_test_data *test_data,
 		const char *arg);
 
@@ -544,8 +555,10 @@  static struct option lgopts[] = {
 	{ CPERF_HUFFMAN_ENC, required_argument, 0, 0 },
 	{ CPERF_LEVEL, required_argument, 0, 0 },
 	{ CPERF_WINDOW_SIZE, required_argument, 0, 0 },
+	{ CPERF_EXTERNAL_MBUFS, 0, 0, 0 },
 	{ NULL, 0, 0, 0 }
 };
+
 static int
 comp_perf_opts_parse_long(int opt_idx, struct comp_test_data *test_data)
 {
@@ -563,6 +576,7 @@  comp_perf_opts_parse_long(int opt_idx, struct comp_test_data *test_data)
 		{ CPERF_HUFFMAN_ENC,	parse_huffman_enc },
 		{ CPERF_LEVEL,		parse_level },
 		{ CPERF_WINDOW_SIZE,	parse_window_sz },
+		{ CPERF_EXTERNAL_MBUFS,	parse_external_mbufs },
 	};
 	unsigned int i;
 
@@ -618,6 +632,7 @@  comp_perf_options_default(struct comp_test_data *test_data)
 	test_data->level_lst.max = RTE_COMP_LEVEL_MAX;
 	test_data->level_lst.inc = 1;
 	test_data->test = CPERF_TEST_TYPE_BENCHMARK;
+	test_data->use_external_mbufs = 0;
 }
 
 int
diff --git a/app/test-compress-perf/comp_perf_test_benchmark.c b/app/test-compress-perf/comp_perf_test_benchmark.c
index 887459449..0c6bb9b45 100644
--- a/app/test-compress-perf/comp_perf_test_benchmark.c
+++ b/app/test-compress-perf/comp_perf_test_benchmark.c
@@ -15,7 +15,8 @@  cperf_benchmark_test_destructor(void *arg)
 {
 	if (arg) {
 		comp_perf_free_memory(
-				&((struct cperf_benchmark_ctx *)arg)->ver.mem);
+			((struct cperf_benchmark_ctx *)arg)->ver.options,
+			&((struct cperf_benchmark_ctx *)arg)->ver.mem);
 		rte_free(arg);
 	}
 }
diff --git a/app/test-compress-perf/comp_perf_test_common.c b/app/test-compress-perf/comp_perf_test_common.c
index 6edc40f04..80683c3a8 100644
--- a/app/test-compress-perf/comp_perf_test_common.c
+++ b/app/test-compress-perf/comp_perf_test_common.c
@@ -77,7 +77,8 @@  find_buf_size(uint32_t input_size)
 }
 
 void
-comp_perf_free_memory(struct cperf_mem_resources *mem)
+comp_perf_free_memory(struct comp_test_data *test_data,
+		      struct cperf_mem_resources *mem)
 {
 	uint32_t i;
 
@@ -93,26 +94,162 @@  comp_perf_free_memory(struct cperf_mem_resources *mem)
 	rte_mempool_free(mem->op_pool);
 	rte_mempool_free(mem->decomp_buf_pool);
 	rte_mempool_free(mem->comp_buf_pool);
+
+	/* external mbuf support */
+	if (mem->decomp_memzones != NULL) {
+		for (i = 0; i < test_data->total_segs; i++)
+			rte_memzone_free(mem->decomp_memzones[i]);
+		rte_free(mem->decomp_memzones);
+	}
+	if (mem->comp_memzones != NULL) {
+		for (i = 0; i < test_data->total_segs; i++)
+			rte_memzone_free(mem->comp_memzones[i]);
+		rte_free(mem->comp_memzones);
+	}
+	rte_free(mem->decomp_buf_infos);
+	rte_free(mem->comp_buf_infos);
+}
+
+static void
+comp_perf_extbuf_free_cb(void *addr __rte_unused, void *opaque __rte_unused)
+{
+}
+
+static const struct rte_memzone *
+comp_perf_make_memzone(const char *name, struct cperf_mem_resources *mem,
+		       unsigned int number, size_t size)
+{
+	unsigned int socket_id = rte_socket_id();
+	char mz_name[RTE_MEMZONE_NAMESIZE];
+	const struct rte_memzone *memzone;
+
+	snprintf(mz_name, RTE_MEMZONE_NAMESIZE, "%s_s%u_d%u_q%u_%d", name,
+		 socket_id, mem->dev_id, mem->qp_id, number);
+	memzone = rte_memzone_lookup(mz_name);
+	if (memzone != NULL && memzone->len != size) {
+		rte_memzone_free(memzone);
+		memzone = NULL;
+	}
+	if (memzone == NULL) {
+		memzone = rte_memzone_reserve_aligned(mz_name, size, socket_id,
+				RTE_MEMZONE_IOVA_CONTIG, RTE_CACHE_LINE_SIZE);
+		if (memzone == NULL)
+			RTE_LOG(ERR, USER1, "Can't allocate memory zone %s\n",
+				mz_name);
+	}
+	return memzone;
+}
+
+static int
+comp_perf_allocate_external_mbufs(struct comp_test_data *test_data,
+				  struct cperf_mem_resources *mem)
+{
+	uint32_t i;
+
+	mem->comp_memzones = rte_zmalloc_socket(NULL,
+		test_data->total_segs * sizeof(struct rte_memzone *),
+		0, rte_socket_id());
+
+	if (mem->comp_memzones == NULL) {
+		RTE_LOG(ERR, USER1,
+			"Memory to hold the compression memzones could not be allocated\n");
+		return -1;
+	}
+
+	mem->decomp_memzones = rte_zmalloc_socket(NULL,
+		test_data->total_segs * sizeof(struct rte_memzone *),
+		0, rte_socket_id());
+
+	if (mem->decomp_memzones == NULL) {
+		RTE_LOG(ERR, USER1,
+			"Memory to hold the decompression memzones could not be allocated\n");
+		return -1;
+	}
+
+	mem->comp_buf_infos = rte_zmalloc_socket(NULL,
+		test_data->total_segs * sizeof(struct rte_mbuf_ext_shared_info),
+		0, rte_socket_id());
+
+	if (mem->comp_buf_infos == NULL) {
+		RTE_LOG(ERR, USER1,
+			"Memory to hold the compression buf infos could not be allocated\n");
+		return -1;
+	}
+
+	mem->decomp_buf_infos = rte_zmalloc_socket(NULL,
+		test_data->total_segs * sizeof(struct rte_mbuf_ext_shared_info),
+		0, rte_socket_id());
+
+	if (mem->decomp_buf_infos == NULL) {
+		RTE_LOG(ERR, USER1,
+			"Memory to hold the decompression buf infos could not be allocated\n");
+		return -1;
+	}
+
+	for (i = 0; i < test_data->total_segs; i++) {
+		mem->comp_memzones[i] = comp_perf_make_memzone("comp", mem,
+				i, test_data->out_seg_sz);
+		if (mem->comp_memzones[i] == NULL) {
+			RTE_LOG(ERR, USER1,
+				"Memory to hold the compression memzone could not be allocated\n");
+			return -1;
+		}
+
+		mem->decomp_memzones[i] = comp_perf_make_memzone("decomp", mem,
+				i, test_data->seg_sz);
+		if (mem->decomp_memzones[i] == NULL) {
+			RTE_LOG(ERR, USER1,
+				"Memory to hold the decompression memzone could not be allocated\n");
+			return -1;
+		}
+
+		mem->comp_buf_infos[i].free_cb =
+				comp_perf_extbuf_free_cb;
+		mem->comp_buf_infos[i].fcb_opaque = NULL;
+		rte_mbuf_ext_refcnt_set(&mem->comp_buf_infos[i], 1);
+
+		mem->decomp_buf_infos[i].free_cb =
+				comp_perf_extbuf_free_cb;
+		mem->decomp_buf_infos[i].fcb_opaque = NULL;
+		rte_mbuf_ext_refcnt_set(&mem->decomp_buf_infos[i], 1);
+	}
+
+	return 0;
 }
 
 int
 comp_perf_allocate_memory(struct comp_test_data *test_data,
 			  struct cperf_mem_resources *mem)
 {
+	uint16_t comp_mbuf_size;
+	uint16_t decomp_mbuf_size;
+
 	test_data->out_seg_sz = find_buf_size(test_data->seg_sz);
+
 	/* Number of segments for input and output
 	 * (compression and decompression)
 	 */
-	uint32_t total_segs = DIV_CEIL(test_data->input_data_sz,
+	test_data->total_segs = DIV_CEIL(test_data->input_data_sz,
 			test_data->seg_sz);
+
+	if (test_data->use_external_mbufs != 0) {
+		if (comp_perf_allocate_external_mbufs(test_data, mem) < 0)
+			return -1;
+		comp_mbuf_size = 0;
+		decomp_mbuf_size = 0;
+	} else {
+		comp_mbuf_size = test_data->out_seg_sz + RTE_PKTMBUF_HEADROOM;
+		decomp_mbuf_size = test_data->seg_sz + RTE_PKTMBUF_HEADROOM;
+	}
+
 	char pool_name[32] = "";
 
 	snprintf(pool_name, sizeof(pool_name), "comp_buf_pool_%u_qp_%u",
 			mem->dev_id, mem->qp_id);
 	mem->comp_buf_pool = rte_pktmbuf_pool_create(pool_name,
-				total_segs,
+				test_data->total_segs,
 				0, 0,
-				test_data->out_seg_sz + RTE_PKTMBUF_HEADROOM,
+				comp_mbuf_size,
 				rte_socket_id());
 	if (mem->comp_buf_pool == NULL) {
 		RTE_LOG(ERR, USER1, "Mbuf mempool could not be created\n");
@@ -122,15 +259,17 @@  comp_perf_allocate_memory(struct comp_test_data *test_data,
 	snprintf(pool_name, sizeof(pool_name), "decomp_buf_pool_%u_qp_%u",
 			mem->dev_id, mem->qp_id);
 	mem->decomp_buf_pool = rte_pktmbuf_pool_create(pool_name,
-				total_segs,
-				0, 0, test_data->seg_sz + RTE_PKTMBUF_HEADROOM,
+				test_data->total_segs,
+				0, 0,
+				decomp_mbuf_size,
 				rte_socket_id());
 	if (mem->decomp_buf_pool == NULL) {
 		RTE_LOG(ERR, USER1, "Mbuf mempool could not be created\n");
 		return -1;
 	}
 
-	mem->total_bufs = DIV_CEIL(total_segs, test_data->max_sgl_segs);
+	mem->total_bufs = DIV_CEIL(test_data->total_segs,
+				   test_data->max_sgl_segs);
 
 	snprintf(pool_name, sizeof(pool_name), "op_pool_%u_qp_%u",
 			mem->dev_id, mem->qp_id);
@@ -148,7 +287,8 @@  comp_perf_allocate_memory(struct comp_test_data *test_data,
 	 */
 	mem->compressed_data = rte_zmalloc_socket(NULL,
 				RTE_MAX(
-				    (size_t) test_data->out_seg_sz * total_segs,
+				    (size_t) test_data->out_seg_sz *
+							  test_data->total_segs,
 				    (size_t) MIN_COMPRESSED_BUF_SIZE),
 				0,
 				rte_socket_id());
@@ -185,7 +325,7 @@  comp_perf_allocate_memory(struct comp_test_data *test_data,
 		return -1;
 	}
 
-	buffer_info.total_segments = total_segs;
+	buffer_info.total_segments = test_data->total_segs;
 	buffer_info.segment_sz = test_data->seg_sz;
 	buffer_info.total_buffs = mem->total_bufs;
 	buffer_info.segments_per_buff = test_data->max_sgl_segs;
@@ -203,6 +343,8 @@  prepare_bufs(struct comp_test_data *test_data, struct cperf_mem_resources *mem)
 	uint8_t *data_addr;
 	uint32_t i, j;
 	uint16_t segs_per_mbuf = 0;
+	uint32_t cmz = 0;
+	uint32_t dmz = 0;
 
 	for (i = 0; i < mem->total_bufs; i++) {
 		/* Allocate data in input mbuf and copy data from input file */
@@ -214,6 +356,16 @@  prepare_bufs(struct comp_test_data *test_data, struct cperf_mem_resources *mem)
 		}
 
 		data_sz = RTE_MIN(remaining_data, test_data->seg_sz);
+
+		if (test_data->use_external_mbufs != 0) {
+			rte_pktmbuf_attach_extbuf(mem->decomp_bufs[i],
+					mem->decomp_memzones[dmz]->addr,
+					mem->decomp_memzones[dmz]->iova,
+					test_data->seg_sz,
+					&mem->decomp_buf_infos[dmz]);
+			dmz++;
+		}
+
 		data_addr = (uint8_t *) rte_pktmbuf_append(
 					mem->decomp_bufs[i], data_sz);
 		if (data_addr == NULL) {
@@ -241,6 +393,17 @@  prepare_bufs(struct comp_test_data *test_data, struct cperf_mem_resources *mem)
 			}
 
 			data_sz = RTE_MIN(remaining_data, test_data->seg_sz);
+
+			if (test_data->use_external_mbufs != 0) {
+				rte_pktmbuf_attach_extbuf(
+					next_seg,
+					mem->decomp_memzones[dmz]->addr,
+					mem->decomp_memzones[dmz]->iova,
+					test_data->seg_sz,
+					&mem->decomp_buf_infos[dmz]);
+				dmz++;
+			}
+
 			data_addr = (uint8_t *)rte_pktmbuf_append(next_seg,
 				data_sz);
 
@@ -268,6 +431,16 @@  prepare_bufs(struct comp_test_data *test_data, struct cperf_mem_resources *mem)
 			RTE_LOG(ERR, USER1, "Could not allocate mbuf\n");
 			return -1;
 		}
+
+		if (test_data->use_external_mbufs != 0) {
+			rte_pktmbuf_attach_extbuf(mem->comp_bufs[i],
+					mem->comp_memzones[cmz]->addr,
+					mem->comp_memzones[cmz]->iova,
+					test_data->out_seg_sz,
+					&mem->comp_buf_infos[cmz]);
+			cmz++;
+		}
+
 		data_addr = (uint8_t *) rte_pktmbuf_append(
 					mem->comp_bufs[i],
 					test_data->out_seg_sz);
@@ -287,6 +460,16 @@  prepare_bufs(struct comp_test_data *test_data, struct cperf_mem_resources *mem)
 				return -1;
 			}
 
+			if (test_data->use_external_mbufs != 0) {
+				rte_pktmbuf_attach_extbuf(
+					next_seg,
+					mem->comp_memzones[cmz]->addr,
+					mem->comp_memzones[cmz]->iova,
+					test_data->out_seg_sz,
+					&mem->comp_buf_infos[cmz]);
+				cmz++;
+			}
+
 			data_addr = (uint8_t *)rte_pktmbuf_append(next_seg,
 				test_data->out_seg_sz);
 			if (data_addr == NULL) {
diff --git a/app/test-compress-perf/comp_perf_test_common.h b/app/test-compress-perf/comp_perf_test_common.h
index c9e0c9081..920642888 100644
--- a/app/test-compress-perf/comp_perf_test_common.h
+++ b/app/test-compress-perf/comp_perf_test_common.h
@@ -26,13 +26,20 @@  struct cperf_mem_resources {
 	struct rte_mempool *comp_buf_pool;
 	struct rte_mempool *decomp_buf_pool;
 	struct rte_mempool *op_pool;
+
+	/* external mbuf support */
+	const struct rte_memzone **comp_memzones;
+	const struct rte_memzone **decomp_memzones;
+	struct rte_mbuf_ext_shared_info *comp_buf_infos;
+	struct rte_mbuf_ext_shared_info *decomp_buf_infos;
 };
 
 int
 param_range_check(uint16_t size, const struct rte_param_log2_range *range);
 
 void
-comp_perf_free_memory(struct cperf_mem_resources *mem);
+comp_perf_free_memory(struct comp_test_data *test_data,
+		      struct cperf_mem_resources *mem);
 
 int
 comp_perf_allocate_memory(struct comp_test_data *test_data,
diff --git a/app/test-compress-perf/comp_perf_test_verify.c b/app/test-compress-perf/comp_perf_test_verify.c
index f66f95ff7..758a22ff5 100644
--- a/app/test-compress-perf/comp_perf_test_verify.c
+++ b/app/test-compress-perf/comp_perf_test_verify.c
@@ -14,7 +14,9 @@  void
 cperf_verify_test_destructor(void *arg)
 {
 	if (arg) {
-		comp_perf_free_memory(&((struct cperf_verify_ctx *)arg)->mem);
+		comp_perf_free_memory(
+				((struct cperf_verify_ctx *)arg)->options,
+				&((struct cperf_verify_ctx *)arg)->mem);
 		rte_free(arg);
 	}
 }