[PATCH] net: davinci emac: use dma_{map, unmap}_single API for cache coherency

[PATCH] net: davinci emac: use dma_{map, unmap}_single API for cache coherency

[Sekhar Nori]

The davinci emac driver uses some ARM specific DMA APIs
for cache coherency which have been removed from kernel
with the 2.6.34 merge.

Modify the driver to use the dma_{map, unmap}_single() APIs
defined in dma-mapping.h

Without this fix, the driver fails to compile on Linus's
tree.

Tested on DM365 and OMAP-L138 EVMs.

Signed-off-by: Sekhar Nori <nsekhar-l0cyMroinI0@public.gmane.org>
---
Applies to latest of Linus's tree.

 drivers/net/davinci_emac.c |   45 +++++++++++++++++++++++++------------------
 1 files changed, 26 insertions(+), 19 deletions(-)

diff --git a/drivers/net/davinci_emac.c b/drivers/net/davinci_emac.c
index 1ac9440..ef0dacf 100644
--- a/drivers/net/davinci_emac.c
+++ b/drivers/net/davinci_emac.c
@@ -29,10 +29,6 @@
  *     PHY layer usage
  */
 
-/** Pending Items in this driver:
- * 1. Use Linux cache infrastcture for DMA'ed memory (dma_xxx functions)
- */
-
 #include <linux/module.h>
 #include <linux/kernel.h>
 #include <linux/sched.h>
@@ -504,12 +500,6 @@ static unsigned long mdio_max_freq;
 
 /* Cache macros - Packet buffers would be from skb pool which is cached */
 #define EMAC_VIRT_NOCACHE(addr) (addr)
-#define EMAC_CACHE_INVALIDATE(addr, size) \
-	dma_cache_maint((void *)addr, size, DMA_FROM_DEVICE)
-#define EMAC_CACHE_WRITEBACK(addr, size) \
-	dma_cache_maint((void *)addr, size, DMA_TO_DEVICE)
-#define EMAC_CACHE_WRITEBACK_INVALIDATE(addr, size) \
-	dma_cache_maint((void *)addr, size, DMA_BIDIRECTIONAL)
 
 /* DM644x does not have BD's in cached memory - so no cache functions */
 #define BD_CACHE_INVALIDATE(addr, size)
@@ -1235,6 +1225,10 @@ static void emac_txch_teardown(struct emac_priv *priv, u32 ch)
 	if (1 == txch->queue_active) {
 		curr_bd = txch->active_queue_head;
 		while (curr_bd != NULL) {
+			dma_unmap_single(emac_dev, curr_bd->buff_ptr,
+				curr_bd->off_b_len & EMAC_RX_BD_BUF_SIZE,
+				DMA_TO_DEVICE);
+
 			emac_net_tx_complete(priv, (void __force *)
 					&curr_bd->buf_token, 1, ch);
 			if (curr_bd != txch->active_queue_tail)
@@ -1327,6 +1321,11 @@ static int emac_tx_bdproc(struct emac_priv *priv, u32 ch, u32 budget)
 				txch->queue_active = 0; /* end of queue */
 			}
 		}
+
+		dma_unmap_single(emac_dev, curr_bd->buff_ptr,
+				curr_bd->off_b_len & EMAC_RX_BD_BUF_SIZE,
+				DMA_TO_DEVICE);
+
 		*tx_complete_ptr = (u32) curr_bd->buf_token;
 		++tx_complete_ptr;
 		++tx_complete_cnt;
@@ -1387,8 +1386,8 @@ static int emac_send(struct emac_priv *priv, struct emac_netpktobj *pkt, u32 ch)
 
 	txch->bd_pool_head = curr_bd->next;
 	curr_bd->buf_token = buf_list->buf_token;
-	/* FIXME buff_ptr = dma_map_single(... data_ptr ...) */
-	curr_bd->buff_ptr = virt_to_phys(buf_list->data_ptr);
+	curr_bd->buff_ptr = dma_map_single(&priv->ndev->dev, buf_list->data_ptr,
+			buf_list->length, DMA_TO_DEVICE);
 	curr_bd->off_b_len = buf_list->length;
 	curr_bd->h_next = 0;
 	curr_bd->next = NULL;
@@ -1468,7 +1467,6 @@ static int emac_dev_xmit(struct sk_buff *skb, struct net_device *ndev)
 	tx_buf.length = skb->len;
 	tx_buf.buf_token = (void *)skb;
 	tx_buf.data_ptr = skb->data;
-	EMAC_CACHE_WRITEBACK((unsigned long)skb->data, skb->len);
 	ndev->trans_start = jiffies;
 	ret_code = emac_send(priv, &tx_packet, EMAC_DEF_TX_CH);
 	if (unlikely(ret_code != 0)) {
@@ -1543,7 +1541,6 @@ static void *emac_net_alloc_rx_buf(struct emac_priv *priv, int buf_size,
 	p_skb->dev = ndev;
 	skb_reserve(p_skb, NET_IP_ALIGN);
 	*data_token = (void *) p_skb;
-	EMAC_CACHE_WRITEBACK_INVALIDATE((unsigned long)p_skb->data, buf_size);
 	return p_skb->data;
 }
 
@@ -1612,8 +1609,8 @@ static int emac_init_rxch(struct emac_priv *priv, u32 ch, char *param)
 		/* populate the hardware descriptor */
 		curr_bd->h_next = emac_virt_to_phys(rxch->active_queue_head,
 				priv);
-		/* FIXME buff_ptr = dma_map_single(... data_ptr ...) */
-		curr_bd->buff_ptr = virt_to_phys(curr_bd->data_ptr);
+		curr_bd->buff_ptr = dma_map_single(emac_dev, curr_bd->data_ptr,
+				rxch->buf_size, DMA_FROM_DEVICE);
 		curr_bd->off_b_len = rxch->buf_size;
 		curr_bd->mode = EMAC_CPPI_OWNERSHIP_BIT;
 
@@ -1697,6 +1694,12 @@ static void emac_cleanup_rxch(struct emac_priv *priv, u32 ch)
 		curr_bd = rxch->active_queue_head;
 		while (curr_bd) {
 			if (curr_bd->buf_token) {
+				dma_unmap_single(&priv->ndev->dev,
+					curr_bd->buff_ptr,
+					curr_bd->off_b_len
+						& EMAC_RX_BD_BUF_SIZE,
+					DMA_FROM_DEVICE);
+
 				dev_kfree_skb_any((struct sk_buff *)\
 						  curr_bd->buf_token);
 			}
@@ -1871,8 +1874,8 @@ static void emac_addbd_to_rx_queue(struct emac_priv *priv, u32 ch,
 
 	/* populate the hardware descriptor */
 	curr_bd->h_next = 0;
-	/* FIXME buff_ptr = dma_map_single(... buffer ...) */
-	curr_bd->buff_ptr = virt_to_phys(buffer);
+	curr_bd->buff_ptr = dma_map_single(&priv->ndev->dev, buffer,
+				rxch->buf_size, DMA_FROM_DEVICE);
 	curr_bd->off_b_len = rxch->buf_size;
 	curr_bd->mode = EMAC_CPPI_OWNERSHIP_BIT;
 	curr_bd->next = NULL;
@@ -1927,7 +1930,6 @@ static int emac_net_rx_cb(struct emac_priv *priv,
 	p_skb = (struct sk_buff *)net_pkt_list->pkt_token;
 	/* set length of packet */
 	skb_put(p_skb, net_pkt_list->pkt_length);
-	EMAC_CACHE_INVALIDATE((unsigned long)p_skb->data, p_skb->len);
 	p_skb->protocol = eth_type_trans(p_skb, priv->ndev);
 	netif_receive_skb(p_skb);
 	priv->net_dev_stats.rx_bytes += net_pkt_list->pkt_length;
@@ -1990,6 +1992,11 @@ static int emac_rx_bdproc(struct emac_priv *priv, u32 ch, u32 budget)
 		rx_buf_obj->data_ptr = (char *)curr_bd->data_ptr;
 		rx_buf_obj->length = curr_bd->off_b_len & EMAC_RX_BD_BUF_SIZE;
 		rx_buf_obj->buf_token = curr_bd->buf_token;
+
+		dma_unmap_single(&priv->ndev->dev, curr_bd->buff_ptr,
+				curr_bd->off_b_len & EMAC_RX_BD_BUF_SIZE,
+				DMA_FROM_DEVICE);
+
 		curr_pkt->pkt_token = curr_pkt->buf_list->buf_token;
 		curr_pkt->num_bufs = 1;
 		curr_pkt->pkt_length =
-- 
1.6.2.4

Re: [PATCH] net: davinci emac: use dma_{map, unmap}_single API for cache coherency

[Kevin Hilman]

Sekhar Nori <nsekhar-l0cyMroinI0@public.gmane.org> writes:

> The davinci emac driver uses some ARM specific DMA APIs
> for cache coherency which have been removed from kernel
> with the 2.6.34 merge.
>
> Modify the driver to use the dma_{map, unmap}_single() APIs
> defined in dma-mapping.h
>
> Without this fix, the driver fails to compile on Linus's
> tree.
>
> Tested on DM365 and OMAP-L138 EVMs.
>
> Signed-off-by: Sekhar Nori <nsekhar-l0cyMroinI0@public.gmane.org>

Acked-by: Kevin Hilman <khilman-1D3HCaltpLuhEniVeURVKkEOCMrvLtNR@public.gmane.org>

Verified that this is compiling/running again with v2.6.34-rc1.

Kevin

Re: [PATCH] net: davinci emac: use dma_{map, unmap}_single API for cache coherency

[David Miller]

From: Kevin Hilman <khilman@deeprootsystems.com>
Date: Thu, 11 Mar 2010 14:56:01 -0800

> Sekhar Nori <nsekhar@ti.com> writes:
> 
>> The davinci emac driver uses some ARM specific DMA APIs
>> for cache coherency which have been removed from kernel
>> with the 2.6.34 merge.
>>
>> Modify the driver to use the dma_{map, unmap}_single() APIs
>> defined in dma-mapping.h
>>
>> Without this fix, the driver fails to compile on Linus's
>> tree.
>>
>> Tested on DM365 and OMAP-L138 EVMs.
>>
>> Signed-off-by: Sekhar Nori <nsekhar@ti.com>
> 
> Acked-by: Kevin Hilman <khilman@deeprootsystems.com>
> 
> Verified that this is compiling/running again with v2.6.34-rc1.

Applied, thanks.