Allocate separate USB buffers for DMA'ed data, so that

DMA data does not reside next to non DMA data. This
might cause more memory to be allocated, but solves
problems on platforms using manual cache
synchronization.

Add a convenience function to get the buffer only
from a USB transfer's page cache structure.

MFC after:	1 week
Suggested by:	imp

1 files changed, 64 insertions, 11 deletions

diff --git a/sys/dev/usb/usb_transfer.c b/sys/dev/usb/usb_transfer.c
index e4cb5fb10502..6e48a3ef9ced 100644
--- a/sys/dev/usb/usb_transfer.c
+++ b/sys/dev/usb/usb_transfer.c
@@ -181,6 +181,10 @@ usbd_get_dma_delay(struct usb_device *udev)
  * according to "size", "align" and "count" arguments. "ppc" is
  * pointed to a linear array of USB page caches afterwards.
  *
+ * If the "align" argument is equal to "1" a non-contiguous allocation
+ * can happen. Else if the "align" argument is greater than "1", the
+ * allocation will always be contiguous in memory.
+ *
  * Returns:
  *    0: Success
  * Else: Failure
@@ -195,13 +199,14 @@ usbd_transfer_setup_sub_malloc(struct usb_setup_params *parm,
 	struct usb_page *pg;
 	void *buf;
 	usb_size_t n_dma_pc;
+	usb_size_t n_dma_pg;
 	usb_size_t n_obj;
 	usb_size_t x;
 	usb_size_t y;
 	usb_size_t r;
 	usb_size_t z;
 
-	USB_ASSERT(align > 1, ("Invalid alignment, 0x%08x\n",
+	USB_ASSERT(align > 0, ("Invalid alignment, 0x%08x\n",
 	    align));
 	USB_ASSERT(size > 0, ("Invalid size = 0\n"));
 
@@ -217,8 +222,14 @@ usbd_transfer_setup_sub_malloc(struct usb_setup_params *parm,
 	 * Try multi-allocation chunks to reduce the number of DMA
 	 * allocations, hence DMA allocations are slow.
 	 */
-	if (size >= USB_PAGE_SIZE) {
+	if (align == 1) {
+		/* special case - non-cached multi page DMA memory */
+		n_dma_pc = count;
+		n_dma_pg = (2 + (size / USB_PAGE_SIZE));
+		n_obj = 1;
+	} else if (size >= USB_PAGE_SIZE) {
 		n_dma_pc = count;
+		n_dma_pg = 1;
 		n_obj = 1;
 	} else {
 		/* compute number of objects per page */
@@ -228,13 +239,22 @@ usbd_transfer_setup_sub_malloc(struct usb_setup_params *parm,
 		 * to nearest one:
 		 */
 		n_dma_pc = ((count + n_obj - 1) / n_obj);
+		n_dma_pg = 1;
 	}
 
+	/*
+	 * DMA memory is allocated once, but mapped twice. That's why
+	 * there is one list for auto-free and another list for
+	 * non-auto-free which only holds the mapping and not the
+	 * allocation.
+	 */
 	if (parm->buf == NULL) {
-		/* for the future */
-		parm->dma_page_ptr += n_dma_pc;
+		/* reserve memory (auto-free) */
+		parm->dma_page_ptr += n_dma_pc * n_dma_pg;
 		parm->dma_page_cache_ptr += n_dma_pc;
-		parm->dma_page_ptr += count;
+
+		/* reserve memory (no-auto-free) */
+		parm->dma_page_ptr += count * n_dma_pg;
 		parm->xfer_page_cache_ptr += count;
 		return (0);
 	}
@@ -272,9 +292,9 @@ usbd_transfer_setup_sub_malloc(struct usb_setup_params *parm,
 		buf = parm->dma_page_cache_ptr->buffer;
 		/* Make room for one DMA page cache and one page */
 		parm->dma_page_cache_ptr++;
-		pg++;
+		pg += n_dma_pg;
 
-		for (y = 0; (y != n_obj); y++, r--, pc++, pg++) {
+		for (y = 0; (y != n_obj); y++, r--, pc++, pg += n_dma_pg) {
 
 			/* Load sub-chunk into DMA */
 			if (usb_pc_dmamap_create(pc, size)) {
@@ -688,12 +708,30 @@ usbd_transfer_setup_sub(struct usb_setup_params *parm)
 	 */
 
 	if (!xfer->flags.ext_buffer) {
+#if USB_HAVE_BUSDMA
+		struct usb_page_search page_info;
+		struct usb_page_cache *pc;
 
+		if (usbd_transfer_setup_sub_malloc(parm,
+		    &pc, parm->bufsize, 1, 1)) {
+			parm->err = USB_ERR_NOMEM;
+		} else if (parm->buf != NULL) {
+
+			usbd_get_page(pc, 0, &page_info);
+
+			xfer->local_buffer = page_info.buffer;
+
+			usbd_xfer_set_frame_offset(xfer, 0, 0);
+
+			if ((type == UE_CONTROL) && (n_frbuffers > 1)) {
+				usbd_xfer_set_frame_offset(xfer, REQ_SIZE, 1);
+			}
+		}
+#else
 		/* align data */
 		parm->size[0] += ((-parm->size[0]) & (USB_HOST_ALIGN - 1));
 
-		if (parm->buf) {
-
+		if (parm->buf != NULL) {
 			xfer->local_buffer =
 			    USB_ADD_BYTES(parm->buf, parm->size[0]);
 
@@ -707,6 +745,7 @@ usbd_transfer_setup_sub(struct usb_setup_params *parm)
 
 		/* align data again */
 		parm->size[0] += ((-parm->size[0]) & (USB_HOST_ALIGN - 1));
+#endif
 	}
 	/*
 	 * Compute maximum buffer size
@@ -1078,9 +1117,12 @@ usbd_transfer_setup(struct usb_device *udev,
 		 * The number of DMA tags required depends on
 		 * the number of endpoints. The current estimate
 		 * for maximum number of DMA tags per endpoint
-		 * is two.
+		 * is three:
+		 * 1) for loading memory
+		 * 2) for allocating memory
+		 * 3) for fixing memory [UHCI]
 		 */
-		parm.dma_tag_max += 2 * MIN(n_setup, USB_EP_MAX);
+		parm.dma_tag_max += 3 * MIN(n_setup, USB_EP_MAX);
 
 		/*
 		 * DMA tags for QH, TD, Data and more.
@@ -1943,6 +1985,17 @@ usbd_xfer_get_frame(struct usb_xfer *xfer, usb_frcount_t frindex)
 	return (&xfer->frbuffers[frindex]);
 }
 
+void *
+usbd_xfer_get_frame_buffer(struct usb_xfer *xfer, usb_frcount_t frindex)
+{
+	struct usb_page_search page_info;
+
+	KASSERT(frindex < xfer->max_frame_count, ("frame index overflow"));
+
+	usbd_get_page(&xfer->frbuffers[frindex], 0, &page_info);
+	return (page_info.buffer);
+}
+
 /*------------------------------------------------------------------------*
  *	usbd_xfer_get_fps_shift
  *