IPEDMA Update

6 files changed, 112 insertions, 37 deletions

diff --git a/dma/ipe.c b/dma/ipe.c
index 63f2fa3..ef3196d 100644
--- a/dma/ipe.c
+++ b/dma/ipe.c
@@ -1,12 +1,14 @@
 #define _PCILIB_DMA_IPE_C
 #define _BSD_SOURCE
 #define _DEFAULT_SOURCE
+#define _POSIX_C_SOURCE 199309L
 
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
 #include <unistd.h>
 #include <sched.h>
+#include <time.h>
 #include <sys/time.h>
 #include <arpa/inet.h>
 
@@ -45,6 +47,10 @@ pcilib_dma_context_t *dma_ipe_init(pcilib_t *pcilib, const char *model, const vo
 	ctx->dma_bank = model_info->banks + dma_bank;
 	ctx->base_addr = pcilib_resolve_register_address(pcilib, ctx->dma_bank->bar, ctx->dma_bank->read_addr);
 
+
+	RD(IPEDMA_REG_VERSION, value);
+	ctx->version = value;
+	
 	RD(IPEDMA_REG_PCIE_GEN, value);
 
 #ifdef IPEDMA_ENFORCE_64BIT_MODE
@@ -101,7 +107,27 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
     if (flags&PCILIB_DMA_FLAG_PERSISTENT) ctx->preserve = 1;
 
     if (ctx->pages) return 0;
-    
+
+    if (!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "dma_timeout", &value))
+	ctx->dma_timeout = value;
+    else 
+	ctx->dma_timeout = IPEDMA_DMA_TIMEOUT;
+
+    if (!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "dma_page_size", &value))
+	ctx->dma_page_size = value;
+    else
+	ctx->dma_page_size = IPEDMA_PAGE_SIZE;
+
+    if (!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "dma_pages", &value))
+	ctx->dma_pages = value;
+    else
+	ctx->dma_pages = IPEDMA_DMA_PAGES;
+
+    if (!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "ipedma_flags", &value))
+	ctx->dma_flags = value;
+    else
+	ctx->dma_flags = 0;
+
     kflags = PCILIB_KMEM_FLAG_REUSE|PCILIB_KMEM_FLAG_EXCLUSIVE|PCILIB_KMEM_FLAG_HARDWARE|(ctx->preserve?PCILIB_KMEM_FLAG_PERSISTENT:0);
     pcilib_kmem_handle_t *desc = pcilib_alloc_kernel_memory(ctx->dmactx.pcilib, PCILIB_KMEM_TYPE_CONSISTENT, 1, IPEDMA_DESCRIPTOR_SIZE, IPEDMA_DESCRIPTOR_ALIGNMENT, PCILIB_KMEM_USE(PCILIB_KMEM_USE_DMA_RING, 0x00), kflags);
     pcilib_kmem_handle_t *pages = pcilib_alloc_kernel_memory(ctx->dmactx.pcilib, PCILIB_KMEM_TYPE_DMA_C2S_PAGE, IPEDMA_DMA_PAGES, 0, 0, PCILIB_KMEM_USE(PCILIB_KMEM_USE_DMA_PAGES, 0x00), kflags);
@@ -409,17 +435,9 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
     volatile uint32_t *empty_detected_ptr;
 
     pcilib_dma_flags_t packet_flags = PCILIB_DMA_FLAG_EOP;
-    size_t nodata_sleep;
 
-    switch (sched_getscheduler(0)) {
-     case SCHED_FIFO:
-     case SCHED_RR:
-	nodata_sleep = IPEDMA_NODATA_SLEEP;
-	break;
-     default:
-	pcilib_info_once("Streaming DMA data using non real-time thread (may cause extra CPU load)", errno);
-	nodata_sleep = 0;
-    }
+    size_t nodata_sleep;
+    struct timespec sleep_ts = {0};
 
     size_t cur_read;
 
@@ -435,6 +453,19 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 
     empty_detected_ptr = last_written_addr_ptr - 2;
 
+    switch (sched_getscheduler(0)) {
+     case SCHED_FIFO:
+     case SCHED_RR:
+        if (ctx->dma_flags&IPEDMA_FLAG_NOSLEEP)
+    	    nodata_sleep = 0;
+    	else
+	    nodata_sleep = IPEDMA_NODATA_SLEEP;
+	break;
+     default:
+	pcilib_info_once("Streaming DMA data using non real-time thread (may cause extra CPU load)", errno);
+	nodata_sleep = 0;
+    }
+
     do {
 	switch (ret&PCILIB_STREAMING_TIMEOUT_MASK) {
 	    case PCILIB_STREAMING_CONTINUE: 
@@ -444,10 +475,10 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 		    wait = 0;
 		else
 #endif /* IPEDMA_SUPPORT_EMPTY_DETECTED */
-		    wait = IPEDMA_DMA_TIMEOUT; 
+		    wait = ctx->dma_timeout; 
 	    break;
 	    case PCILIB_STREAMING_WAIT: 
-		wait = (timeout > IPEDMA_DMA_TIMEOUT)?timeout:IPEDMA_DMA_TIMEOUT;
+		wait = (timeout > ctx->dma_timeout)?timeout:ctx->dma_timeout;
 	    break;
 //	    case PCILIB_STREAMING_CHECK: wait = 0; break;
 	}
@@ -460,8 +491,10 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 	gettimeofday(&start, NULL);
 	memcpy(&cur, &start, sizeof(struct timeval));
 	while (((*last_written_addr_ptr == 0)||(ctx->last_read_addr == (*last_written_addr_ptr)))&&((wait == PCILIB_TIMEOUT_INFINITE)||(((cur.tv_sec - start.tv_sec)*1000000 + (cur.tv_usec - start.tv_usec)) < wait))) {
-	    if (nodata_sleep) 
-		usleep(nodata_sleep);
+	    if (nodata_sleep) {
+	        sleep_ts.tv_nsec = nodata_sleep;
+		nanosleep(&sleep_ts, NULL);
+	    }
 		
 #ifdef IPEDMA_SUPPORT_EMPTY_DETECTED
 	    if ((ret != PCILIB_STREAMING_REQ_PACKET)&&(*empty_detected_ptr)) break;
@@ -494,13 +527,14 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 	else packet_flags = 0;
 #endif /* IPEDMA_DETECT_PACKETS */
 	
-	pcilib_kmem_sync_block(ctx->dmactx.pcilib, ctx->pages, PCILIB_KMEM_SYNC_FROMDEVICE, cur_read);
+	if ((ctx->dma_flags&IPEDMA_FLAG_NOSYNC) == 0)
+	    pcilib_kmem_sync_block(ctx->dmactx.pcilib, ctx->pages, PCILIB_KMEM_SYNC_FROMDEVICE, cur_read);
         void *buf = pcilib_kmem_get_block_ua(ctx->dmactx.pcilib, ctx->pages, cur_read);
 	ret = cb(cbattr, packet_flags, ctx->page_size, buf);
 	if (ret < 0) return -ret;
 	
 	    // We don't need this because hardware does not intend to read anything from the memory
-//	pcilib_kmem_sync_block(ctx->dmactx.pcilib, ctx->pages, PCILIB_KMEM_SYNC_TODEVICE, cur_read);
+	//pcilib_kmem_sync_block(ctx->dmactx.pcilib, ctx->pages, PCILIB_KMEM_SYNC_TODEVICE, cur_read);
 
 	    // Return buffer into the DMA pool when processed
 	if (ctx->streaming) {
@@ -510,7 +544,7 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 	    else last_free = IPEDMA_DMA_PAGES - 1;
 
 	    uintptr_t buf_ba = pcilib_kmem_get_block_ba(ctx->dmactx.pcilib, ctx->pages, last_free);
-	    WR(IPEDMA_REG_PAGE_ADDR, buf_ba);    
+	    WR(IPEDMA_REG_PAGE_ADDR, buf_ba);
 # ifdef IPEDMA_STREAMING_CHECKS
 	    pcilib_register_value_t streaming_status;
 	    RD(IPEDMA_REG_STREAMING_STATUS, streaming_status);
diff --git a/dma/ipe.h b/dma/ipe.h
index ce3ea4f..174e36e 100644
--- a/dma/ipe.h
+++ b/dma/ipe.h
@@ -5,6 +5,11 @@
 #include "pcilib.h"
 #include "version.h"
 
+#define IPEDMA_PAGE_SIZE		4096		/**< page size */
+#define IPEDMA_DMA_PAGES		1024		/**< number of DMA pages in the ring buffer to allocate */
+
+#define IPEDMA_DMA_TIMEOUT 		100000		/**< us, overrides PCILIB_DMA_TIMEOUT (actual hardware timeout is 50ms according to Lorenzo) */
+
 pcilib_dma_context_t *dma_ipe_init(pcilib_t *ctx, const char *model, const void *arg);
 void  dma_ipe_free(pcilib_dma_context_t *vctx);
 
@@ -68,6 +73,7 @@ static const pcilib_register_description_t ipe_dma_registers[] = {
     {0x0018, 	0, 	32, 	0, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "dma_mode_flags",		"DMA operation mode"},
     {0x0018, 	0, 	4, 	0, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_BITS, PCILIB_REGISTER_BANK_DMA, "pcie_gen",  			"PCIe version 2/3 depending on the used XILINX core"},
     {0x0018, 	4, 	1, 	0, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_BITS, PCILIB_REGISTER_BANK_DMA, "streaming_dma", 			"Streaming mode (enabled/disabled)"},
+    {0x0020,	0,	32,	0,	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "dma_firmware",		"Version of DMA firmware"},
     {0x0028, 	0, 	32, 	0, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "mwr_perf",			"MWR Performance"},
     {0x003C, 	0, 	32, 	0, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "cfg_lnk_width",		"Negotiated and max width of PCIe Link"},
     {0x003C, 	0, 	6, 	0, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_BITS, PCILIB_REGISTER_BANK_DMA, "cfg_cap_max_lnk_width", 		"Max link width"},
@@ -81,9 +87,18 @@ static const pcilib_register_description_t ipe_dma_registers[] = {
     {0x0058, 	0, 	32, 	0, 	0x00000000,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "last_descriptor_read",	"Last descriptor read by the host"},
     {0x005C, 	0, 	32, 	0, 	0x00000000,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "desc_mem_addr", 		"Number of descriptors configured"},
     {0x0060, 	0, 	32, 	0, 	0x00000000,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMA, "update_thresh",  		"Update threshold of progress register"},
-    {0x0000, 	0, 	32, 	PCILIB_VERSION, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMACONF, "dma_version",  		"Version of DMA engine"},
+    {0x0000, 	0, 	32, 	PCILIB_VERSION, 	0x00000000,	PCILIB_REGISTER_R   , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMACONF, "dma_version",	"Version of DMA engine"},
+    {0x0004, 	0, 	32, 	IPEDMA_DMA_TIMEOUT, 	0x00000000,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMACONF, "dma_timeout",	"Default DMA timeout"},
+    {0x0008, 	0, 	32, 	IPEDMA_DMA_PAGES,	0x00000000,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMACONF, "dma_pages",	"Number of buffers in DMA page ring"},
+    {0x000C, 	0, 	32, 	IPEDMA_PAGE_SIZE,	0x00000000,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMACONF, "dma_page_size",	"Size of a page in DMA page ring (multiple of 4K)"},
+    {0x0010, 	0, 	32, 	0,			0xFFFFFFFF,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_STANDARD, PCILIB_REGISTER_BANK_DMACONF, "ipedma_flags",	"DMA Control Register"},
+    {0x0010, 	0, 	1, 	0,			0xFFFFFFFF,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_BITS,	PCILIB_REGISTER_BANK_DMACONF, "ipedma_nosync",	"Do not synchronize DMA pages"},
+    {0x0010, 	1, 	1, 	0,			0xFFFFFFFF,	PCILIB_REGISTER_RW  , PCILIB_REGISTER_BITS,	PCILIB_REGISTER_BANK_DMACONF, "ipedma_nosleep",	"Do not sleep while there is no data"},
     {0,		0,	0,	0,	0x00000000,	0,                                           0,                        0, NULL, 			NULL}
 };
+
+
+
 #endif /* _PCILIB_EXPORT_C */
 
 
diff --git a/dma/ipe_benchmark.c b/dma/ipe_benchmark.c
index fa89f4b..57e5646 100644
--- a/dma/ipe_benchmark.c
+++ b/dma/ipe_benchmark.c
@@ -1,6 +1,7 @@
 #define _PCILIB_DMA_IPE_C
 #define _BSD_SOURCE
 #define _DEFAULT_SOURCE
+#define _POSIX_C_SOURCE 200112L
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -84,6 +85,7 @@ double dma_ipe_benchmark(pcilib_dma_context_t *vctx, pcilib_dma_engine_addr_t dm
 	read_dma = pcilib_read_dma_custom;
 
 	// There is no significant difference and we can remove this when testing phase is over.
+	// DS: With large number of buffers this is quite slow due to skimming of initially written buffers
     if (getenv("PCILIB_BENCHMARK_STREAMING")) {
 	size_t dma_buffer_space;
 	pcilib_dma_engine_status_t dma_status;
@@ -97,7 +99,7 @@ double dma_ipe_benchmark(pcilib_dma_context_t *vctx, pcilib_dma_engine_addr_t dm
 	WR(IPEDMA_REG_CONTROL, 0x1);
 
 	gettimeofday(&start, NULL);
-	pcilib_calc_deadline(&start, IPEDMA_DMA_TIMEOUT * IPEDMA_DMA_PAGES);
+	pcilib_calc_deadline(&start, ctx->dma_timeout * IPEDMA_DMA_PAGES);
 
 #ifdef IPEDMA_BUG_LAST_READ
 	dma_buffer_space = (IPEDMA_DMA_PAGES - 2) * IPEDMA_PAGE_SIZE;
@@ -106,8 +108,8 @@ double dma_ipe_benchmark(pcilib_dma_context_t *vctx, pcilib_dma_engine_addr_t dm
 #endif /* IPEDMA_BUG_LAST_READ */
 
 	// Allocate memory and prepare data
-	buf = malloc(size + dma_buffer_space);
-	if (!buf) return -1;
+	err = posix_memalign(&buf, 4096, size + dma_buffer_space);
+	if ((err)||(!buf)) return -1;
 
 	    // Wait all DMA buffers are filled 
 	memset(&dma_status, 0, sizeof(dma_status));
@@ -127,7 +129,7 @@ double dma_ipe_benchmark(pcilib_dma_context_t *vctx, pcilib_dma_engine_addr_t dm
 	gettimeofday(&start, NULL);
 	for (iter = 0; iter < iterations; iter++) {
 	    for (bytes = 0; bytes < (size + dma_buffer_space); bytes += rbytes) {
-		err = read_dma(ctx->dmactx.pcilib, 0, addr, size + dma_buffer_space - bytes, PCILIB_DMA_FLAG_MULTIPACKET, PCILIB_DMA_TIMEOUT,  buf + bytes, &rbytes);
+		err = read_dma(ctx->dmactx.pcilib, 0, addr, size + dma_buffer_space - bytes, PCILIB_DMA_FLAG_MULTIPACKET, ctx->dma_timeout,  buf + bytes, &rbytes);
 	        if (err) {
 		    pcilib_error("Can't read data from DMA, error %i", err);
 		    return -1;
@@ -141,12 +143,15 @@ double dma_ipe_benchmark(pcilib_dma_context_t *vctx, pcilib_dma_engine_addr_t dm
 
 	    // Stopping DMA
 	WR(IPEDMA_REG_CONTROL, 0x0);
+	usleep(IPEDMA_RESET_DELAY);
+	
 	pcilib_skip_dma(ctx->dmactx.pcilib, 0);
     } else {
 	if (read_dma == dma_ipe_skim_dma_custom)
 	    pcilib_info_once("Benchmarking the DMA hardware (without memcpy)");
 
 	WR(IPEDMA_REG_CONTROL, 0x0);
+	usleep(IPEDMA_RESET_DELAY);
 
 	err = pcilib_skip_dma(ctx->dmactx.pcilib, 0);
 	if (err) {
@@ -155,35 +160,44 @@ double dma_ipe_benchmark(pcilib_dma_context_t *vctx, pcilib_dma_engine_addr_t dm
 	}
 
 	// Allocate memory and prepare data
-	buf = malloc(size);
-	if (!buf) return -1;
+	err = posix_memalign(&buf, 4096, size);
+	if ((err)||(!buf)) return -1;
 
-	for (iter = 0; iter < iterations; iter++) {
+	for (iter = 0; iter <= iterations; iter++) {
 	    gettimeofday(&start, NULL);
 
 	    // Starting DMA
 	    WR(IPEDMA_REG_CONTROL, 0x1);
 	
 	    for (bytes = 0; bytes < size; bytes += rbytes) {
-		err = read_dma(ctx->dmactx.pcilib, 0, addr, size - bytes, PCILIB_DMA_FLAG_MULTIPACKET, PCILIB_DMA_TIMEOUT,  buf + bytes, &rbytes);
+		err = read_dma(ctx->dmactx.pcilib, 0, addr, size - bytes, PCILIB_DMA_FLAG_MULTIPACKET, ctx->dma_timeout,  buf + bytes, &rbytes);
 	        if (err) {
 		    pcilib_error("Can't read data from DMA (iteration: %zu, offset: %zu), error %i", iter, bytes, err);
 		    return -1;
 		}
 	    }
 
+	    gettimeofday(&cur, NULL);
+
 		// Stopping DMA
 	    WR(IPEDMA_REG_CONTROL, 0x0);
+	    usleep(IPEDMA_RESET_DELAY);
 	    if (err) break;
 	
-	    gettimeofday(&cur, NULL);
-	    us += ((cur.tv_sec - start.tv_sec)*1000000 + (cur.tv_usec - start.tv_usec));
-	    
+		// Heating up during the first iteration
+	    if (iter)
+		us += ((cur.tv_sec - start.tv_sec)*1000000 + (cur.tv_usec - start.tv_usec));
+
+	    pcilib_info("Iteration %-4i latency: %lu", iter, ((cur.tv_sec - start.tv_sec)*1000000 + (cur.tv_usec - start.tv_usec)));
+
+
 	    err = pcilib_skip_dma(ctx->dmactx.pcilib, 0);
 	    if (err) {
 		pcilib_error("Can't start iteration, devices continuously writes unexpected data using DMA engine");
 		break;
 	    }
+	    
+	    usleep(ctx->dma_timeout);
 
 	}
     }
diff --git a/dma/ipe_private.h b/dma/ipe_private.h
index 5054a58..fdc2d83 100644
--- a/dma/ipe_private.h
+++ b/dma/ipe_private.h
@@ -10,8 +10,6 @@
 
 #define IPEDMA_STREAMING_MODE				/**< Enables streaming DMA operation mode instead of ring-buffer, the page is written once and forgotten and need to be pushed in queue again */
 //#define IPEDMA_STREAMING_CHECKS				/**< Enables status checks in streaming mode (it will cause _significant_ performance penalty, max ~ 2 GB/s) */
-#define IPEDMA_PAGE_SIZE		4096
-#define IPEDMA_DMA_PAGES		1024		/**< number of DMA pages in the ring buffer to allocate */
 #define IPEDMA_DMA_PROGRESS_THRESHOLD	1		/**< how many pages the DMA engine should fill before reporting progress */
 #define IPEDMA_DESCRIPTOR_SIZE		128
 #define IPEDMA_DESCRIPTOR_ALIGNMENT	64
@@ -21,16 +19,13 @@
 //#define IPEDMA_DETECT_PACKETS				/**< Using empty_deceted flag */
 #define  IPEDMA_SUPPORT_EMPTY_DETECTED			/**< Avoid waiting for data when empty_detected flag is set in hardware */
 
-#define IPEDMA_DMA_TIMEOUT 		100000		/**< us, overrides PCILIB_DMA_TIMEOUT (actual hardware timeout is 50ms according to Lorenzo) */
-#define IPEDMA_RESET_DELAY		100000		/**< Sleep between accessing DMA control and reset registers */
-#define IPEDMA_ADD_PAGE_DELAY		1000		/**< Delay between submitting successive DMA pages into IPEDMA_REG_PAGE_ADDR register */
-#define IPEDMA_NODATA_SLEEP		10		/**< To keep CPU free */
 
 #define IPEDMA_REG_RESET		0x00
 #define IPEDMA_REG_CONTROL		0x04
 #define IPEDMA_REG_TLP_SIZE		0x0C
 #define IPEDMA_REG_TLP_COUNT		0x10
 #define IPEDMA_REG_PCIE_GEN		0x18
+#define IPEDMA_REG_VERSION		0x20
 #define IPEDMA_REG_PAGE_ADDR		0x50
 #define IPEDMA_REG_UPDATE_ADDR		0x54
 #define IPEDMA_REG_LAST_READ		0x58		/**< In streaming mode, we can use it freely to track current status */
@@ -38,9 +33,16 @@
 #define IPEDMA_REG_UPDATE_THRESHOLD	0x60
 #define IPEDMA_REG_STREAMING_STATUS	0x68
 
+#define IPEDMA_FLAG_NOSYNC		0x01		/**< Do not call kernel space for page synchronization */
+#define IPEDMA_FLAG_NOSLEEP		0x02		/**< Do not sleep in the loop while waiting for the data */
+
 #define IPEDMA_MASK_PCIE_GEN		0xF
 #define IPEDMA_MASK_STREAMING_MODE	0x10
 
+#define IPEDMA_RESET_DELAY		10000		/**< Sleep between accessing DMA control and reset registers */
+#define IPEDMA_ADD_PAGE_DELAY		1000		/**< Delay between submitting successive DMA pages into IPEDMA_REG_PAGE_ADDR register */
+#define IPEDMA_NODATA_SLEEP		100		/**< To keep CPU free, in nanoseconds */
+
 
 #define WR(addr, value) { *(uint32_t*)(ctx->base_addr + addr) = value; }
 #define RD(addr, value) { value = *(uint32_t*)(ctx->base_addr + addr); }
@@ -59,6 +61,7 @@ struct ipe_dma_s {
     pcilib_irq_type_t irq_preserve;	/**< indicates that IRQs should not be disabled during clean-up */
     int irq_started;			/**< indicates that IRQ subsystem is initialized (detecting which types should be preserverd) */    
 
+    uint32_t version;			/**< hardware version */
     int started;			/**< indicates that DMA buffers are initialized and reading is allowed */
     int writting;			/**< indicates that we are in middle of writting packet */
     int reused;				/**< indicates that DMA was found intialized, buffers were reused, and no additional initialization is needed */
@@ -66,6 +69,11 @@ struct ipe_dma_s {
     int mode64;				/**< indicates 64-bit operation mode */
     int streaming;			/**< indicates if DMA is operating in streaming or ring-buffer mode */
 
+    uint32_t dma_flags;			/**< Various operation flags, see IPEDMA_FLAG_* */
+    size_t dma_timeout;			/**< DMA timeout,IPEDMA_DMA_TIMEOUT is used by default */
+    size_t dma_pages;			/**< Number of DMA pages in ring buffer to allocate */
+    size_t dma_page_size;		/**< Size of a single DMA page */
+
     pcilib_kmem_handle_t *desc;		/**< in-memory status descriptor written by DMA engine upon operation progess */
     pcilib_kmem_handle_t *pages;	/**< collection of memory-locked pages for DMA operation */
 
diff --git a/pcilib/property.c b/pcilib/property.c
index 09ed449..a7d1a61 100644
--- a/pcilib/property.c
+++ b/pcilib/property.c
@@ -72,7 +72,7 @@ int pcilib_add_properties_from_registers(pcilib_t *ctx, size_t n, const pcilib_r
         char *view_name;
         pcilib_access_mode_t mode = 0;
 
-        pcilib_register_view_description_t v = {0};
+        pcilib_register_view_description_t v = {{0}};
         pcilib_register_bank_description_t *b = &ctx->banks[banks[i]];
 
         if (registers[i].type == PCILIB_REGISTER_PROPERTY) continue;
diff --git a/pcitool/cli.c b/pcitool/cli.c
index 0696ef4..90f11d4 100644
--- a/pcitool/cli.c
+++ b/pcitool/cli.c
@@ -3715,6 +3715,10 @@ int main(int argc, char **argv) {
         if (amode != ACCESS_DMA)
 	    break;
      case MODE_BENCHMARK:
+        sched_param.sched_priority = sched_get_priority_max(SCHED_FIFO);
+        err = sched_setscheduler(0, SCHED_FIFO, &sched_param);
+        if (err) pcilib_info("Failed to acquire real-time priority (errno: %i)", errno);
+     break;
      case MODE_GRAB:
         sched_param.sched_priority = sched_get_priority_min(SCHED_FIFO);
         err = sched_setscheduler(0, SCHED_FIFO, &sched_param);