Configure number of DMA buffers in IPEDMA and improve checking and reporting inconsistent kmem buffers while re-using

6 files changed, 193 insertions, 112 deletions

diff --git a/dma/ipe.c b/dma/ipe.c
index 320955c..2c3f32c 100644
--- a/dma/ipe.c
+++ b/dma/ipe.c
@@ -110,6 +110,24 @@ void  dma_ipe_free(pcilib_dma_context_t *vctx) {
 }
 
 
+static void dma_ipe_disable(ipe_dma_t *ctx) {
+	    // Disable DMA
+    WR(IPEDMA_REG_CONTROL, 0x0);
+    usleep(IPEDMA_RESET_DELAY);
+	
+	    // Reset DMA engine
+    WR(IPEDMA_REG_RESET, 0x1);
+    usleep(IPEDMA_RESET_DELAY);
+    WR(IPEDMA_REG_RESET, 0x0);
+    usleep(IPEDMA_RESET_DELAY);
+
+        // Reseting configured DMA pages
+    if (ctx->version < 3) {
+	WR(IPEDMA_REG2_PAGE_COUNT, 0);
+    }
+    usleep(IPEDMA_RESET_DELAY);
+}
+
 int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dma_flags_t flags) {
     int err;
     int mask = 32;
@@ -169,10 +187,10 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
     } else
 	ctx->page_size = IPEDMA_PAGE_SIZE;
 
-    if (!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "dma_pages", &value))
-	ctx->dma_pages = value;
+    if ((!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "dma_pages", &value))&&(value > 0))
+	ctx->ring_size = value;
     else
-	ctx->dma_pages = IPEDMA_DMA_PAGES;
+	ctx->ring_size = IPEDMA_DMA_PAGES;
 
     if (!pcilib_read_register(ctx->dmactx.pcilib, "dmaconf", "ipedma_flags", &value))
 	ctx->dma_flags = value;
@@ -191,35 +209,56 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
 
     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, ctx->page_size, 0, PCILIB_KMEM_USE(PCILIB_KMEM_USE_DMA_PAGES, 0x00), kflags);
+    pcilib_kmem_handle_t *pages = pcilib_alloc_kernel_memory(ctx->dmactx.pcilib, PCILIB_KMEM_TYPE_DMA_C2S_PAGE, ctx->ring_size, ctx->page_size, 0, PCILIB_KMEM_USE(PCILIB_KMEM_USE_DMA_PAGES, 0x00), kflags);
 
     if (!desc||!pages) {
-	if (pages) pcilib_free_kernel_memory(ctx->dmactx.pcilib, pages, 0);
-	if (desc) pcilib_free_kernel_memory(ctx->dmactx.pcilib, desc, 0);
+	if (pages) pcilib_free_kernel_memory(ctx->dmactx.pcilib, pages, KMEM_FLAG_REUSE);
+	if (desc) pcilib_free_kernel_memory(ctx->dmactx.pcilib, desc, KMEM_FLAG_REUSE);
+	printf("%lu\n", IPEDMA_DESCRIPTOR_SIZE);
+	pcilib_error("Can't allocate required kernel memory for IPEDMA engine (%lu pages of %lu bytes + %lu byte descriptor)", ctx->ring_size, ctx->page_size, (unsigned long)IPEDMA_DESCRIPTOR_SIZE);
 	return PCILIB_ERROR_MEMORY;
     }
     reuse_desc = pcilib_kmem_is_reused(ctx->dmactx.pcilib, desc);
     reuse_pages = pcilib_kmem_is_reused(ctx->dmactx.pcilib, pages);
 
-    if (reuse_desc == reuse_pages) {
-	if (reuse_desc & PCILIB_KMEM_REUSE_PARTIAL) pcilib_warning("Inconsistent DMA buffers are found (only part of required buffers is available), reinitializing...");
-	else if (reuse_desc & PCILIB_KMEM_REUSE_REUSED) {
-	    if ((reuse_desc & PCILIB_KMEM_REUSE_PERSISTENT) == 0) pcilib_warning("Lost DMA buffers are found (non-persistent mode), reinitializing...");
-	    else if ((reuse_desc & PCILIB_KMEM_REUSE_HARDWARE) == 0) pcilib_warning("Lost DMA buffers are found (missing HW reference), reinitializing...");
+    if ((reuse_pages & PCILIB_KMEM_REUSE_PARTIAL)||(reuse_desc & PCILIB_KMEM_REUSE_PARTIAL)) {
+	dma_ipe_disable(ctx);
+
+	pcilib_free_kernel_memory(ctx->dmactx.pcilib, pages, KMEM_FLAG_REUSE);
+	pcilib_free_kernel_memory(ctx->dmactx.pcilib, desc, KMEM_FLAG_REUSE);
+
+	if ((flags&PCILIB_DMA_FLAG_STOP) == 0) {
+	    pcilib_error("Inconsistent DMA buffers are found (buffers are only partially re-used). This is very wrong, please stop DMA engine and correct configuration...");
+	    return PCILIB_ERROR_INVALID_STATE;
+	}
+	
+	pcilib_warning("Inconsistent DMA buffers are found (buffers are only partially re-used), reinitializing...");
+	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_FLAG_MASS);
+	pages = pcilib_alloc_kernel_memory(ctx->dmactx.pcilib, PCILIB_KMEM_TYPE_DMA_C2S_PAGE, ctx->ring_size, ctx->page_size, 0, PCILIB_KMEM_USE(PCILIB_KMEM_USE_DMA_PAGES, 0x00), kflags|PCILIB_KMEM_FLAG_MASS);
+
+	if (!desc||!pages) {
+	    if (pages) pcilib_free_kernel_memory(ctx->dmactx.pcilib, pages, KMEM_FLAG_REUSE);
+	    if (desc) pcilib_free_kernel_memory(ctx->dmactx.pcilib, desc, KMEM_FLAG_REUSE);
+	    return PCILIB_ERROR_MEMORY;
+	}
+    } else if (reuse_desc != reuse_pages) {
+	 pcilib_warning("Inconsistent DMA buffers (modes of ring and page buffers does not match), reinitializing....");
+    } else if (reuse_desc & PCILIB_KMEM_REUSE_REUSED) {
+	if ((reuse_desc & PCILIB_KMEM_REUSE_PERSISTENT) == 0) pcilib_warning("Lost DMA buffers are found (non-persistent mode), reinitializing...");
+	else if ((reuse_desc & PCILIB_KMEM_REUSE_HARDWARE) == 0) pcilib_warning("Lost DMA buffers are found (missing HW reference), reinitializing...");
+	else {
+	    if (ctx->streaming)
+		preserve = 1;
 	    else {
-		if (ctx->streaming)
+		RD(IPEDMA_REG2_PAGE_COUNT, value);
+
+		if (value != ctx->ring_size) 
+		    pcilib_warning("Inconsistent DMA buffers are found (Number of allocated buffers (%lu) does not match current request (%lu)), reinitializing...", value + 1, IPEDMA_DMA_PAGES);
+		else 
 		    preserve = 1;
-		else {
-		    RD(IPEDMA_REG2_PAGE_COUNT, value);
-
-		    if (value != IPEDMA_DMA_PAGES) 
-			pcilib_warning("Inconsistent DMA buffers are found (Number of allocated buffers (%lu) does not match current request (%lu)), reinitializing...", value + 1, IPEDMA_DMA_PAGES);
-		    else 
-			preserve = 1;
-		} 
-	    }
+	    } 
 	}
-    } else pcilib_warning("Inconsistent DMA buffers (modes of ring and page buffers does not match), reinitializing....");
+    }
 
     desc_va = pcilib_kmem_get_ua(ctx->dmactx.pcilib, desc);
     if (ctx->version < 3) {
@@ -242,7 +281,7 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
 	RD(ctx->reg_last_read, value);
 	    // Numbered from 1 in FPGA
 # ifdef IPEDMA_BUG_LAST_READ
-	if (value == IPEDMA_DMA_PAGES)
+	if (value == ctx->ring_size)
 	    value = 0;
 # else /* IPEDMA_BUG_LAST_READ */
 	value--;
@@ -251,16 +290,8 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
 	ctx->last_read = value;
     } else {
 	ctx->reused = 0;
-
-	    // Disable DMA
-	WR(IPEDMA_REG_CONTROL, 0x0);
-	usleep(IPEDMA_RESET_DELAY);
 	
-	    // Reset DMA engine
-	WR(IPEDMA_REG_RESET, 0x1);
-	usleep(IPEDMA_RESET_DELAY);
-	WR(IPEDMA_REG_RESET, 0x0);
-	usleep(IPEDMA_RESET_DELAY);
+	dma_ipe_disable(ctx);
 
 	    // Verify PCIe link status
 	RD(IPEDMA_REG_RESET, value);
@@ -284,9 +315,9 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
 
 	    // Setting current read position and configuring progress register
 #ifdef IPEDMA_BUG_LAST_READ
-	WR(ctx->reg_last_read, IPEDMA_DMA_PAGES - 1);
+	WR(ctx->reg_last_read, ctx->ring_size - 1);
 #else /* IPEDMA_BUG_LAST_READ */
-	WR(ctx->reg_last_read, IPEDMA_DMA_PAGES);
+	WR(ctx->reg_last_read, ctx->ring_size);
 #endif /* IPEDMA_BUG_LAST_READ */
 
 	if (ctx->version < 3) {
@@ -304,7 +335,7 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
 	    // In ring buffer mode, the hardware taking care to preserve an empty buffer to help distinguish between
 	    // completely empty and completely full cases. In streaming mode, it is our responsibility to track this
 	    // information. Therefore, we always keep the last buffer free
-	num_pages = IPEDMA_DMA_PAGES;
+	num_pages = ctx->ring_size;
 	if (ctx->streaming) num_pages--;
 	
 	for (i = 0; i < num_pages; i++) {
@@ -330,7 +361,7 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
 	    // Enable DMA
 	WR(IPEDMA_REG_CONTROL, 0x1);
 	
-	ctx->last_read = IPEDMA_DMA_PAGES - 1;
+	ctx->last_read = ctx->ring_size - 1;
     }
 
     ctx->last_read_addr = pcilib_kmem_get_block_ba(ctx->dmactx.pcilib, pages, ctx->last_read);
@@ -338,8 +369,6 @@ int dma_ipe_start(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dm
     ctx->desc = desc;
     ctx->pages = pages;
 
-    ctx->ring_size = IPEDMA_DMA_PAGES;
-
     return 0;
 }
 
@@ -364,22 +393,7 @@ int dma_ipe_stop(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, pcilib_dma
 
 	ctx->started  = 0;
 
-	    // Disable DMA
-	WR(IPEDMA_REG_CONTROL, 0);
-	usleep(IPEDMA_RESET_DELAY);
-	
-	    // Reset DMA engine
-	WR(IPEDMA_REG_RESET, 0x1);
-	usleep(IPEDMA_RESET_DELAY);
-	WR(IPEDMA_REG_RESET, 0x0);
-	usleep(IPEDMA_RESET_DELAY);
-
-	    // Reseting configured DMA pages
-	if (ctx->version < 3) {
-	    WR(IPEDMA_REG2_PAGE_COUNT, 0);
-	}
-
-	usleep(IPEDMA_RESET_DELAY);
+	dma_ipe_disable(ctx);
     }
 
 	// Clean buffers
@@ -633,7 +647,7 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 	    size_t last_free;
 		// We always keep 1 buffer free to distinguish between completely full and empty cases
 	    if (cur_read) last_free = cur_read - 1;
-	    else last_free = IPEDMA_DMA_PAGES - 1;
+	    else last_free = ctx->ring_size - 1;
 
 	    uintptr_t buf_ba = pcilib_kmem_get_block_ba(ctx->dmactx.pcilib, ctx->pages, last_free);
 	    if (ctx->version < 3) {
@@ -651,7 +665,7 @@ int dma_ipe_stream_read(pcilib_dma_context_t *vctx, pcilib_dma_engine_t dma, uin
 
 	    // Numbered from 1
 #ifdef IPEDMA_BUG_LAST_READ
-	WR(ctx->reg_last_read, cur_read?cur_read:IPEDMA_DMA_PAGES);
+	WR(ctx->reg_last_read, cur_read?cur_read:ctx->ring_size);
 #else /* IPEDMA_BUG_LAST_READ */
 	WR(ctx->reg_last_read, cur_read + 1);
 #endif /* IPEDMA_BUG_LAST_READ */
diff --git a/driver/ioctl.c b/driver/ioctl.c
index 0814fc8..26618c0 100644
--- a/driver/ioctl.c
+++ b/driver/ioctl.c
@@ -181,15 +181,13 @@ static int ioctl_pci_info(pcidriver_privdata_t *privdata, unsigned long arg)
  */
 static int ioctl_kmem_alloc(pcidriver_privdata_t *privdata, unsigned long arg)
 {
-	int ret;
-	READ_FROM_USER(kmem_handle_t, khandle);
-
-	if ((ret = pcidriver_kmem_alloc(privdata, &khandle)) != 0)
-		return ret;
+	int err, ret;
 
+	READ_FROM_USER(kmem_handle_t, khandle);
+	err = pcidriver_kmem_alloc(privdata, &khandle);
 	WRITE_TO_USER(kmem_handle_t, khandle);
 
-	return 0;
+	return err;
 }
 
 /**
diff --git a/driver/kmem.c b/driver/kmem.c
index 7539ae6..618b5bf 100644
--- a/driver/kmem.c
+++ b/driver/kmem.c
@@ -53,23 +53,27 @@ int pcidriver_kmem_alloc(pcidriver_privdata_t *privdata, kmem_handle_t *kmem_han
 		} else {
 		    if (kmem_handle->type != kmem_entry->type) {
 		    	mod_info("Invalid type of reusable kmem_entry, currently: %lu, but requested: %lu\n", kmem_entry->type, kmem_handle->type);
+		    	kmem_handle->type = kmem_entry->type;
 			return -EINVAL;
 		    }
 
 		    if (((kmem_handle->type&PCILIB_KMEM_TYPE_MASK) == PCILIB_KMEM_TYPE_PAGE)&&(kmem_handle->size == 0)) {
-			    kmem_handle->size = kmem_entry->size;
+			kmem_handle->size = kmem_entry->size;
 		    } else if (kmem_handle->size != kmem_entry->size) {
 			mod_info("Invalid size of reusable kmem_entry, currently: %lu, but requested: %lu\n", kmem_entry->size, kmem_handle->size);
+			kmem_handle->size = kmem_entry->size;
 			return -EINVAL;
 		    }
-		    
+
 		    if (kmem_handle->align != kmem_entry->align) {
 			mod_info("Invalid alignment of reusable kmem_entry, currently: %lu, but requested: %lu\n", kmem_entry->align, kmem_handle->align);
+			kmem_handle->align = kmem_entry->align;
 			return -EINVAL;
 		    }
 
 		    if (((kmem_entry->mode&KMEM_MODE_EXCLUSIVE)?1:0) != ((flags&KMEM_FLAG_EXCLUSIVE)?1:0)) {
 			mod_info("Invalid mode of reusable kmem_entry\n");
+			kmem_handle->flags = (kmem_entry->mode&KMEM_MODE_EXCLUSIVE)?KMEM_FLAG_EXCLUSIVE:0;
 			return -EINVAL;
 		    }
 		}
diff --git a/pcilib/kmem.c b/pcilib/kmem.c
index b1d2c5c..4e240c3 100644
--- a/pcilib/kmem.c
+++ b/pcilib/kmem.c
@@ -46,8 +46,8 @@ static void pcilib_cancel_kernel_memory(pcilib_t *ctx, pcilib_kmem_list_t *kbuf,
     if (last_flags) {
 	pcilib_kmem_flags_t failed_flags = flags;
 	
-	if (last_flags&KMEM_FLAG_REUSED_PERSISTENT) flags&=~PCILIB_KMEM_FLAG_PERSISTENT;
-	if (last_flags&KMEM_FLAG_REUSED_HW) flags&=~PCILIB_KMEM_FLAG_HARDWARE;
+	if (last_flags&KMEM_FLAG_REUSED_PERSISTENT) failed_flags&=~PCILIB_KMEM_FLAG_PERSISTENT;
+	if (last_flags&KMEM_FLAG_REUSED_HW) failed_flags&=~PCILIB_KMEM_FLAG_HARDWARE;
 	
 	if (failed_flags != flags) {
 	    ret = pcilib_free_kernel_buffer(ctx, kbuf, --kbuf->buf.n_blocks, failed_flags);
@@ -60,10 +60,10 @@ static void pcilib_cancel_kernel_memory(pcilib_t *ctx, pcilib_kmem_list_t *kbuf,
 
 pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type_t type, size_t nmemb, size_t size, size_t alignment, pcilib_kmem_use_t use, pcilib_kmem_flags_t flags) {
     int err = 0;
-    const char *error = NULL;
+    char error[256];
     
     int ret;
-    int i;
+    size_t i, allocated = nmemb;
     void *addr;
     
     pcilib_tristate_t reused = PCILIB_TRISTATE_NO;
@@ -92,7 +92,7 @@ pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type
 	pcilib_error("PCIDRIVER_IOC_MMAP_MODE ioctl have failed");
 	return NULL;
     }
-    
+
     kh.type = type;
     kh.size = size;
     kh.align = alignment;
@@ -104,10 +104,13 @@ pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type
 	kh.size += alignment;
     }
 
-    for ( i = 0; i < nmemb; i++) {
+    for ( i = 0; (i < nmemb)||(flags&PCILIB_KMEM_FLAG_MASS); i++) {
 	kh.item = i;
 	kh.flags = flags;
 
+	if (i >= nmemb) 
+	    kh.flags |= KMEM_FLAG_TRY;
+
 	if ((type&PCILIB_KMEM_TYPE_MASK) == PCILIB_KMEM_TYPE_REGION) {
 	    kh.pa = alignment + i * size;
 	}
@@ -115,10 +118,43 @@ pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type
         ret = ioctl(ctx->handle, PCIDRIVER_IOC_KMEM_ALLOC, &kh);
 	if (ret) {
 	    kbuf->buf.n_blocks = i;
-	    error = "PCIDRIVER_IOC_KMEM_ALLOC ioctl have failed";
+	    if ((i < nmemb)||(errno != ENOENT)) {
+		err = PCILIB_ERROR_FAILED;
+		if (errno == EINVAL) {
+		    if (kh.type != type)
+			sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  we have requested type %u but buffer is of type %lu", use, i, type, kh.type);
+		    else if (kh.size != size)
+			sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  we have requested size %lu but buffer is of size %lu", use, i, size, kh.size);
+		    else if (kh.align != alignment)
+			sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  we have requested alignment %lu but buffer is of alignment %lu", use, i, size, kh.size);
+		    else if ((kh.flags&KMEM_FLAG_EXCLUSIVE) != (flags&KMEM_FLAG_EXCLUSIVE))
+			sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  we have requested size %s but buffer is of size %s", use, i, ((flags&KMEM_FLAG_EXCLUSIVE)?"exclusive":"non-exclusive"), ((kh.flags&KMEM_FLAG_EXCLUSIVE)?"exclusive":"non exclusive"));
+		    else 
+			sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  unknown consistency error", use, i);
+		} else if (errno == EBUSY) {
+		    sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  reuse counter of kmem_entry is overflown", use, i);
+		} else if (errno == ENOMEM) {
+		    sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  memory allocation (%zu bytes) failed", use, i, size);
+		} else {
+		    sprintf(error, "Driver prevents us from re-using buffer (use 0x%x, block: %zu),  PCIDRIVER_IOC_KMEM_ALLOC ioctl have failed with errno %i", use, i, errno);
+		}
+	    }
 	    break;
 	}
 	
+	if (i >= allocated) {
+	    void *kbuf_new = realloc(kbuf, sizeof(pcilib_kmem_list_t) + 2 * allocated * sizeof(pcilib_kmem_addr_t));
+	    if (!kbuf_new) {
+		kbuf->buf.n_blocks = i;
+		err = PCILIB_ERROR_MEMORY;
+		sprintf(error, "Failed to allocate extra %zu bytes of user-space memory for kmem structures", allocated * sizeof(pcilib_kmem_addr_t));
+		break;
+	    }
+	    memset(kbuf_new + sizeof(pcilib_kmem_list_t) + allocated * sizeof(pcilib_kmem_addr_t) , 0, allocated * sizeof(pcilib_kmem_addr_t));
+	    kbuf = kbuf_new;
+	    allocated *= 2;
+	}
+	
 	kbuf->buf.blocks[i].handle_id = kh.handle_id;
 	kbuf->buf.blocks[i].pa = kh.pa;
 	kbuf->buf.blocks[i].size = kh.size;
@@ -135,27 +171,44 @@ pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type
 		    else*/ persistent = (kh.flags&KMEM_FLAG_REUSED_PERSISTENT)?1:0;
 		} else if ((kh.flags&KMEM_FLAG_REUSED_PERSISTENT) == 0) err = PCILIB_ERROR_INVALID_STATE;
 	    } else if (kh.flags&KMEM_FLAG_REUSED_PERSISTENT) err = PCILIB_ERROR_INVALID_STATE;
-	    
+	    if (err) {
+		kbuf->buf.n_blocks = i + 1;
+		sprintf(error,  "Mistmatch in persistent modes of the re-used kmem blocks. Current buffer (use 0x%x, block: %zu) is %s, but prior ones %s", 
+		    use, i, ((kh.flags&KMEM_FLAG_REUSED_PERSISTENT)?"persistent":"not persistent"), (persistent?"are":"are not"));
+		break;
+	    }
+
 	    if (hardware) {
 		if (hardware < 0) {
 		    /*if (((flags&PCILIB_KMEM_FLAG_HARDWARE) == 0)&&(kh.flags&KMEM_FLAG_REUSED_HW)) err = PCILIB_ERROR_INVALID_STATE;
 		    else*/ hardware = (kh.flags&KMEM_FLAG_REUSED_HW)?1:0;
 		} else if ((kh.flags&KMEM_FLAG_REUSED_HW) == 0) err = PCILIB_ERROR_INVALID_STATE;
 	    } else if (kh.flags&KMEM_FLAG_REUSED_HW) err = PCILIB_ERROR_INVALID_STATE;
-	    
+	    if (err) {
+		kbuf->buf.n_blocks = i + 1;
+		sprintf(error,  "Mistmatch in hardware modes of the re-used kmem blocks. Current buffer (use 0x%x, block: %zu) is %s, but prior ones %s", 
+		    use, i, ((kh.flags&KMEM_FLAG_REUSED_HW)?"hardware-locked":"not hardware-locked"), (hardware?"are":"are not"));
+		break;
+	    }
 	} else {
 	    if (!i) reused = PCILIB_TRISTATE_NO;
 	    else if (reused) reused = PCILIB_TRISTATE_PARTIAL;
-	    
-	    if ((persistent > 0)&&((flags&PCILIB_KMEM_FLAG_PERSISTENT) == 0)) err = PCILIB_ERROR_INVALID_STATE;
-	    if ((hardware > 0)&&((flags&PCILIB_KMEM_FLAG_HARDWARE) == 0)) err = PCILIB_ERROR_INVALID_STATE;
+
+	    if ((persistent > 0)&&((flags&PCILIB_KMEM_FLAG_PERSISTENT) == 0)) {
+		err = PCILIB_ERROR_INVALID_STATE;
+		sprintf(error, "Expecting to re-use persistent blocks, but buffer (use 0x%x, block: %zu) is not", use, i);
+	    }
+	    else if ((hardware > 0)&&((flags&PCILIB_KMEM_FLAG_HARDWARE) == 0)) {
+		err = PCILIB_ERROR_INVALID_STATE;
+		sprintf(error, "Expecting to re-use hardware-locked blocks, but buffer (use 0x%x, block: %zu) is not", use, i);
+	    }
+	    if (err) {
+		kbuf->buf.n_blocks = i + 1;
+		break;
+	    }
 	}
 	
-	if (err) {
-	    kbuf->buf.n_blocks = i + 1;
-	    break;
-	}
-    
+
         if ((kh.align)&&((kh.type&PCILIB_KMEM_TYPE_MASK) != PCILIB_KMEM_TYPE_PAGE)) {
 	    if (kh.pa % kh.align) kbuf->buf.blocks[i].alignment_offset = kh.align - kh.pa % kh.align;
 	    kbuf->buf.blocks[i].size -= kh.align;
@@ -164,18 +217,34 @@ pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type
     	addr = mmap( 0, kbuf->buf.blocks[i].size + kbuf->buf.blocks[i].alignment_offset, PROT_WRITE | PROT_READ, MAP_SHARED, ctx->handle, 0 );
 	if ((!addr)||(addr == MAP_FAILED)) {
 	    kbuf->buf.n_blocks = i + 1;
-	    error = "Failed to mmap allocated kernel memory";
+	    err = PCILIB_ERROR_FAILED;
+	    sprintf(error, "Driver prevents us from mmaping buffer (use 0x%x, block: %zu), mmap have failed with errno %i", use, i, errno);
 	    break;
 	}
 
 	kbuf->buf.blocks[i].ua = addr;
-//	if (use == PCILIB_KMEM_USE_DMA_PAGES) {
-//	memset(addr, 10, kbuf->buf.blocks[i].size + kbuf->buf.blocks[i].alignment_offset);
-//	}
-	
 	kbuf->buf.blocks[i].mmap_offset = kh.pa & ctx->page_mask;
     }
 
+    if (err) kbuf->buf.n_blocks = i + 1;
+    else kbuf->buf.n_blocks = i;
+
+
+	// Check if there are more unpicked buffers
+    if ((!err)&&((flags&PCILIB_KMEM_FLAG_MASS) == 0)&&(reused == PCILIB_TRISTATE_YES)&&((type&PCILIB_KMEM_TYPE_MASK) != PCILIB_KMEM_TYPE_REGION)) {
+	kh.item = kbuf->buf.n_blocks;
+	kh.flags = KMEM_FLAG_REUSE|KMEM_FLAG_TRY;
+
+        ret = ioctl(ctx->handle, PCIDRIVER_IOC_KMEM_ALLOC, &kh);
+	if (!ret) {
+	    kh.flags = KMEM_FLAG_REUSE;
+	    ioctl(ctx->handle, PCIDRIVER_IOC_KMEM_FREE, &kh);
+	    reused = PCILIB_TRISTATE_PARTIAL;
+	} else if (errno != ENOENT) {
+	    reused = PCILIB_TRISTATE_PARTIAL;
+	}
+    }
+
     pcilib_unlock_global(ctx);
 
 
@@ -183,37 +252,32 @@ pcilib_kmem_handle_t *pcilib_alloc_kernel_memory(pcilib_t *ctx, pcilib_kmem_type
     if (persistent < 0) persistent = 0;
     if (hardware < 0) hardware = 0;
 
-    if (err||error) {
-	pcilib_kmem_flags_t free_flags = 0;
-	
-	    // for the sake of simplicity always clean partialy reused buffers
-	if ((persistent == PCILIB_TRISTATE_PARTIAL)||((persistent <= 0)&&(flags&PCILIB_KMEM_FLAG_PERSISTENT))) {
-	    free_flags |= PCILIB_KMEM_FLAG_PERSISTENT;
-	}
-	
-	if ((hardware <= 0)&&(flags&PCILIB_KMEM_FLAG_HARDWARE)) {
-	    free_flags |= PCILIB_KMEM_FLAG_HARDWARE;
-	}
-	
-	    // do not clean if we have reused peresistent buffers
-	    //  we don't care about -1, because it will be the value only if no buffers actually allocated
-	if ((!persistent)||(reused != PCILIB_TRISTATE_YES)) {
-	    pcilib_cancel_kernel_memory(ctx, kbuf, free_flags, err?kh.flags:0);
+    if (err) {
+	    // do not clean if we have reused (even partially) persistent/hardware-locked buffers
+	if (((persistent)||(hardware))&&(reused != PCILIB_TRISTATE_NO)) {
+	    pcilib_cancel_kernel_memory(ctx, kbuf, KMEM_FLAG_REUSE, 0);
+	} else {
+	    pcilib_kmem_flags_t free_flags = 0;
+	    if (flags&PCILIB_KMEM_FLAG_PERSISTENT) {
+		free_flags |= PCILIB_KMEM_FLAG_PERSISTENT;
+	    }
+	    if (flags&PCILIB_KMEM_FLAG_HARDWARE) {
+		free_flags |= PCILIB_KMEM_FLAG_HARDWARE;
+	    }
+		// err indicates consistensy error. The last ioctl have succeeded and we need to clean it in a special way
+	    pcilib_cancel_kernel_memory(ctx, kbuf, free_flags, (err == PCILIB_ERROR_INVALID_STATE)?kh.flags:0);
 	}
 
-	if (!error) error = "Reused buffers are inconsistent";
-	pcilib_error(error);
-
+	pcilib_warning("Error %i: %s", err, error);
 	return NULL;
     }
     
     if (nmemb == 1) {
 	memcpy(&kbuf->buf.addr, &kbuf->buf.blocks[0], sizeof(pcilib_kmem_addr_t));
     }
-    
+
     kbuf->buf.reused = reused|(persistent?PCILIB_KMEM_REUSE_PERSISTENT:0)|(hardware?PCILIB_KMEM_REUSE_HARDWARE:0);
-    kbuf->buf.n_blocks = nmemb;
-    
+
     kbuf->prev = NULL;
     kbuf->next = ctx->kmem_list;
     if (ctx->kmem_list) ctx->kmem_list->prev = kbuf;
diff --git a/pcilib/pcilib.h b/pcilib/pcilib.h
index fbab883..cfe3e96 100644
--- a/pcilib/pcilib.h
+++ b/pcilib/pcilib.h
@@ -86,7 +86,8 @@ typedef enum {
     PCILIB_DMA_FLAG_WAIT = 2,			/**< wait completion of write operation / wait for data during read operation */
     PCILIB_DMA_FLAG_MULTIPACKET = 4,		/**< read multiple packets */
     PCILIB_DMA_FLAG_PERSISTENT = 8,		/**< do not stop DMA engine on application termination / permanently close DMA engine on dma_stop */
-    PCILIB_DMA_FLAG_IGNORE_ERRORS = 16		/**< do not crash on errors, but return appropriate error codes */
+    PCILIB_DMA_FLAG_IGNORE_ERRORS = 16,		/**< do not crash on errors, but return appropriate error codes */
+    PCILIB_DMA_FLAG_STOP = 32			/**< indicates that we actually calling pcilib_dma_start to stop persistent DMA engine */
 } pcilib_dma_flags_t;
 
 typedef enum {
diff --git a/pcitool/cli.c b/pcitool/cli.c
index b54ba9d..0a070d7 100644
--- a/pcitool/cli.c
+++ b/pcitool/cli.c
@@ -2333,7 +2333,7 @@ int StartStopDMA(pcilib_t *handle,  const pcilib_model_description_t *model_info
         if (start) Error("DMA engine should be specified");
 
 	for (dmaid = 0; dma_info->engines[dmaid].addr_bits; dmaid++) {
-	    err = pcilib_start_dma(handle, dmaid, 0);
+	    err = pcilib_start_dma(handle, dmaid, PCILIB_DMA_FLAG_STOP);
 	    if (err) Error("Error starting DMA Engine (%s %i)", ((dma_info->engines[dmaid].direction == PCILIB_DMA_FROM_DEVICE)?"C2S":"S2C"), dma_info->engines[dmaid].addr);
 	    err = pcilib_stop_dma(handle, dmaid, PCILIB_DMA_FLAG_PERSISTENT);
 	    if (err) Error("Error stopping DMA Engine (%s %i)", ((dma_info->engines[dmaid].direction == PCILIB_DMA_FROM_DEVICE)?"C2S":"S2C"), dma_info->engines[dmaid].addr);
@@ -2350,7 +2350,7 @@ int StartStopDMA(pcilib_t *handle,  const pcilib_model_description_t *model_info
 	    err = pcilib_start_dma(handle, dmaid, PCILIB_DMA_FLAG_PERSISTENT);
     	    if (err) Error("Error starting DMA engine (C2S %lu)", dma);
 	} else {
-	    err = pcilib_start_dma(handle, dmaid, 0);
+	    err = pcilib_start_dma(handle, dmaid, PCILIB_DMA_FLAG_STOP);
     	    if (err) Error("Error starting DMA engine (C2S %lu)", dma);
 	    err = pcilib_stop_dma(handle, dmaid, PCILIB_DMA_FLAG_PERSISTENT);
     	    if (err) Error("Error stopping DMA engine (C2S %lu)", dma);
@@ -2365,7 +2365,7 @@ int StartStopDMA(pcilib_t *handle,  const pcilib_model_description_t *model_info
 	    err = pcilib_start_dma(handle, dmaid, PCILIB_DMA_FLAG_PERSISTENT);
     	    if (err) Error("Error starting DMA engine (S2C %lu)", dma);
 	} else {
-	    err = pcilib_start_dma(handle, dmaid, 0);
+	    err = pcilib_start_dma(handle, dmaid, PCILIB_DMA_FLAG_STOP);
     	    if (err) Error("Error starting DMA engine (S2C %lu)", dma);
 	    err = pcilib_stop_dma(handle, dmaid, PCILIB_DMA_FLAG_PERSISTENT);
     	    if (err) Error("Error stopping DMA engine (S2C %lu)", dma);