[PATCH 0/5] qlge: Fixes from PPC bringup.

[PATCH 0/5] qlge: Fixes from PPC bringup.

[Ron Mercer]

Dave/Jeff,

The attached patch series replaces those I sent on 12/24.  Patch 3/5 is
cleaned up per Ben and Christoph.  The rest have the file mode changed
removed per Dave.

Christoph also asked me to verify the entire driver for endianess with
sparse which I am now working on.  In the meantime I would like to get
these changes dropped in if possible.

Regards,
Ron

[PATCH 1/5] qlge: bugfix: Add missing pci_mapping_err checking.

[Ron Mercer]

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index 718a7bd..c6ab6a4 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -963,6 +963,11 @@ static void ql_update_sbq(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 						     sbq_desc->p.skb->data,
 						     rx_ring->sbq_buf_size /
 						     2, PCI_DMA_FROMDEVICE);
+				if (pci_dma_mapping_error(qdev->pdev, map)) {
+					QPRINTK(qdev, IFUP, ERR, "PCI mapping failed.\n");
+					rx_ring->sbq_clean_idx = clean_idx;
+					return;
+				}
 				pci_unmap_addr_set(sbq_desc, mapaddr, map);
 				pci_unmap_len_set(sbq_desc, maplen,
 						  rx_ring->sbq_buf_size / 2);
-- 
1.6.0

Re: [PATCH 1/5] qlge: bugfix: Add missing pci_mapping_err checking.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Tue, 30 Dec 2008 13:12:21 -0800

> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

[PATCH 2/5] qlge: bugfix: Add missing pci_unmap_page call in receive path.

[Ron Mercer]

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index c6ab6a4..9ceedfc 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -1308,6 +1308,11 @@ static struct sk_buff *ql_build_rx_skb(struct ql_adapter *qdev,
 					"No skb available, drop the packet.\n");
 				return NULL;
 			}
+			pci_unmap_page(qdev->pdev,
+				       pci_unmap_addr(lbq_desc,
+						      mapaddr),
+				       pci_unmap_len(lbq_desc, maplen),
+				       PCI_DMA_FROMDEVICE);
 			skb_reserve(skb, NET_IP_ALIGN);
 			QPRINTK(qdev, RX_STATUS, DEBUG,
 				"%d bytes of headers and data in large. Chain page to new skb and pull tail.\n", length);
-- 
1.6.0

Re: [PATCH 2/5] qlge: bugfix: Add missing pci_unmap_page call in receive path.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Tue, 30 Dec 2008 13:12:22 -0800

> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

[PATCH 3/5] qlge: bugfix: Fix shadow register endian issue.

[Ron Mercer]

Shadow registers are consistent memory locations to which the chip
echos ring indexes in little endian format.  These values need to
be endian swapped before referencing.

Note:
The register pointer declaration uses the volatile modifier which
causes warnings in checkpatch.
Per Documentation/volatile-considered-harmful.txt:
  - Pointers to data structures in coherent memory which might be modified
    by I/O devices can, sometimes, legitimately be volatile.  A ring buffer
    used by a network adapter, where that adapter changes pointers to
    indicate which descriptors have been processed, is an example of this
    type of situation.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge.h      |   17 +----------------
 drivers/net/qlge/qlge_main.c |   12 ++++++------
 2 files changed, 7 insertions(+), 22 deletions(-)

diff --git a/drivers/net/qlge/qlge.h b/drivers/net/qlge/qlge.h
index ba2e1c5..97321bb 100644
--- a/drivers/net/qlge/qlge.h
+++ b/drivers/net/qlge/qlge.h
@@ -1189,7 +1189,7 @@ struct rx_ring {
 	u32 cq_size;
 	u32 cq_len;
 	u16 cq_id;
-	u32 *prod_idx_sh_reg;	/* Shadowed producer register. */
+	volatile __le32 *prod_idx_sh_reg;	/* Shadowed producer register. */
 	dma_addr_t prod_idx_sh_reg_dma;
 	void __iomem *cnsmr_idx_db_reg;	/* PCI doorbell mem area + 0 */
 	u32 cnsmr_idx;		/* current sw idx */
@@ -1467,21 +1467,6 @@ static inline void ql_write_db_reg(u32 val, void __iomem *addr)
 	mmiowb();
 }
 
-/*
- * Shadow Registers:
- * Outbound queues have a consumer index that is maintained by the chip.
- * Inbound queues have a producer index that is maintained by the chip.
- * For lower overhead, these registers are "shadowed" to host memory
- * which allows the device driver to track the queue progress without
- * PCI reads. When an entry is placed on an inbound queue, the chip will
- * update the relevant index register and then copy the value to the
- * shadow register in host memory.
- */
-static inline unsigned int ql_read_sh_reg(const volatile void  *addr)
-{
-	return *(volatile unsigned int __force *)addr;
-}
-
 extern char qlge_driver_name[];
 extern const char qlge_driver_version[];
 extern const struct ethtool_ops qlge_ethtool_ops;
diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index 9ceedfc..c0e43c5 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -1559,7 +1559,7 @@ static void ql_process_chip_ae_intr(struct ql_adapter *qdev,
 static int ql_clean_outbound_rx_ring(struct rx_ring *rx_ring)
 {
 	struct ql_adapter *qdev = rx_ring->qdev;
-	u32 prod = ql_read_sh_reg(rx_ring->prod_idx_sh_reg);
+	u32 prod = le32_to_cpu(*rx_ring->prod_idx_sh_reg);
 	struct ob_mac_iocb_rsp *net_rsp = NULL;
 	int count = 0;
 
@@ -1585,7 +1585,7 @@ static int ql_clean_outbound_rx_ring(struct rx_ring *rx_ring)
 		}
 		count++;
 		ql_update_cq(rx_ring);
-		prod = ql_read_sh_reg(rx_ring->prod_idx_sh_reg);
+		prod = le32_to_cpu(*rx_ring->prod_idx_sh_reg);
 	}
 	ql_write_cq_idx(rx_ring);
 	if (netif_queue_stopped(qdev->ndev) && net_rsp != NULL) {
@@ -1605,7 +1605,7 @@ static int ql_clean_outbound_rx_ring(struct rx_ring *rx_ring)
 static int ql_clean_inbound_rx_ring(struct rx_ring *rx_ring, int budget)
 {
 	struct ql_adapter *qdev = rx_ring->qdev;
-	u32 prod = ql_read_sh_reg(rx_ring->prod_idx_sh_reg);
+	u32 prod = le32_to_cpu(*rx_ring->prod_idx_sh_reg);
 	struct ql_net_rsp_iocb *net_rsp;
 	int count = 0;
 
@@ -1638,7 +1638,7 @@ static int ql_clean_inbound_rx_ring(struct rx_ring *rx_ring, int budget)
 		}
 		count++;
 		ql_update_cq(rx_ring);
-		prod = ql_read_sh_reg(rx_ring->prod_idx_sh_reg);
+		prod = le32_to_cpu(*rx_ring->prod_idx_sh_reg);
 		if (count == budget)
 			break;
 	}
@@ -1801,7 +1801,7 @@ static irqreturn_t qlge_isr(int irq, void *dev_id)
 	 * Check the default queue and wake handler if active.
 	 */
 	rx_ring = &qdev->rx_ring[0];
-	if (ql_read_sh_reg(rx_ring->prod_idx_sh_reg) != rx_ring->cnsmr_idx) {
+	if (le32_to_cpu(*rx_ring->prod_idx_sh_reg) != rx_ring->cnsmr_idx) {
 		QPRINTK(qdev, INTR, INFO, "Waking handler for rx_ring[0].\n");
 		ql_disable_completion_interrupt(qdev, intr_context->intr);
 		queue_delayed_work_on(smp_processor_id(), qdev->q_workqueue,
@@ -1815,7 +1815,7 @@ static irqreturn_t qlge_isr(int irq, void *dev_id)
 		 */
 		for (i = 1; i < qdev->rx_ring_count; i++) {
 			rx_ring = &qdev->rx_ring[i];
-			if (ql_read_sh_reg(rx_ring->prod_idx_sh_reg) !=
+			if (le32_to_cpu(*rx_ring->prod_idx_sh_reg) !=
 			    rx_ring->cnsmr_idx) {
 				QPRINTK(qdev, INTR, INFO,
 					"Waking handler for rx_ring[%d].\n", i);
-- 
1.6.0

Re: [PATCH 3/5] qlge: bugfix: Fix shadow register endian issue.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Tue, 30 Dec 2008 13:12:23 -0800

> Shadow registers are consistent memory locations to which the chip
> echos ring indexes in little endian format.  These values need to
> be endian swapped before referencing.
> 
> Note:
> The register pointer declaration uses the volatile modifier which
> causes warnings in checkpatch.
> Per Documentation/volatile-considered-harmful.txt:
>   - Pointers to data structures in coherent memory which might be modified
>     by I/O devices can, sometimes, legitimately be volatile.  A ring buffer
>     used by a network adapter, where that adapter changes pointers to
>     indicate which descriptors have been processed, is an example of this
>     type of situation.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied, but still kill the volatile and use proper memory
barriers around accesses to this value instead.

Many other drivers handle exactly this situation without
having to resort to volatile.

[PATCH 4/5] qlge: bugfix: Fix ring length setting for rx ring, large/small

[Ron Mercer]

The length field for these rings is 16-bits.  If the length is
the max supported 65536 then the setting should be zero.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |   13 +++++++++----
 1 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index c0e43c5..ffa2100 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -2491,7 +2491,8 @@ static int ql_start_rx_ring(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 	memset((void *)cqicb, 0, sizeof(struct cqicb));
 	cqicb->msix_vect = rx_ring->irq;
 
-	cqicb->len = cpu_to_le16(rx_ring->cq_len | LEN_V | LEN_CPP_CONT);
+	bq_len = (rx_ring->cq_len == 65536) ? 0 : (u16) rx_ring->cq_len;
+	cqicb->len = cpu_to_le16(bq_len | LEN_V | LEN_CPP_CONT);
 
 	cqicb->addr_lo = cpu_to_le32(rx_ring->cq_base_dma);
 	cqicb->addr_hi = cpu_to_le32((u64) rx_ring->cq_base_dma >> 32);
@@ -2513,8 +2514,11 @@ static int ql_start_rx_ring(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		    cpu_to_le32(rx_ring->lbq_base_indirect_dma);
 		cqicb->lbq_addr_hi =
 		    cpu_to_le32((u64) rx_ring->lbq_base_indirect_dma >> 32);
-		cqicb->lbq_buf_size = cpu_to_le32(rx_ring->lbq_buf_size);
-		bq_len = (u16) rx_ring->lbq_len;
+		bq_len = (rx_ring->lbq_buf_size == 65536) ? 0 :
+			(u16) rx_ring->lbq_buf_size;
+		cqicb->lbq_buf_size = cpu_to_le16(bq_len);
+		bq_len = (rx_ring->lbq_len == 65536) ? 0 :
+			(u16) rx_ring->lbq_len;
 		cqicb->lbq_len = cpu_to_le16(bq_len);
 		rx_ring->lbq_prod_idx = rx_ring->lbq_len - 16;
 		rx_ring->lbq_curr_idx = 0;
@@ -2530,7 +2534,8 @@ static int ql_start_rx_ring(struct ql_adapter *qdev, struct rx_ring *rx_ring)
 		    cpu_to_le32((u64) rx_ring->sbq_base_indirect_dma >> 32);
 		cqicb->sbq_buf_size =
 		    cpu_to_le16(((rx_ring->sbq_buf_size / 2) + 8) & 0xfffffff8);
-		bq_len = (u16) rx_ring->sbq_len;
+		bq_len = (rx_ring->sbq_len == 65536) ? 0 :
+			(u16) rx_ring->sbq_len;
 		cqicb->sbq_len = cpu_to_le16(bq_len);
 		rx_ring->sbq_prod_idx = rx_ring->sbq_len - 16;
 		rx_ring->sbq_curr_idx = 0;
-- 
1.6.0

Re: [PATCH 4/5] qlge: bugfix: Fix ring length setting for rx ring, large/small

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Tue, 30 Dec 2008 13:12:24 -0800

> The length field for these rings is 16-bits.  If the length is
> the max supported 65536 then the setting should be zero.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Applied.

[PATCH 5/5] qlge: bugfix: Fix register access error checking.

[Ron Mercer]

Some indexed registers do not have error bits. In these cases a
value of zero should be used for error checking.

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |   24 ++++++++++++------------
 1 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index ffa2100..837be72 100644
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -257,7 +257,7 @@ int ql_get_mac_addr_reg(struct ql_adapter *qdev, u32 type, u16 index,
 		{
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 			if (status)
 				goto exit;
 			ql_write32(qdev, MAC_ADDR_IDX, (offset++) | /* offset */
@@ -265,13 +265,13 @@ int ql_get_mac_addr_reg(struct ql_adapter *qdev, u32 type, u16 index,
 				   MAC_ADDR_ADR | MAC_ADDR_RS | type); /* type */
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MR, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MR, 0);
 			if (status)
 				goto exit;
 			*value++ = ql_read32(qdev, MAC_ADDR_DATA);
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 			if (status)
 				goto exit;
 			ql_write32(qdev, MAC_ADDR_IDX, (offset++) | /* offset */
@@ -279,14 +279,14 @@ int ql_get_mac_addr_reg(struct ql_adapter *qdev, u32 type, u16 index,
 				   MAC_ADDR_ADR | MAC_ADDR_RS | type); /* type */
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MR, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MR, 0);
 			if (status)
 				goto exit;
 			*value++ = ql_read32(qdev, MAC_ADDR_DATA);
 			if (type == MAC_ADDR_TYPE_CAM_MAC) {
 				status =
 				    ql_wait_reg_rdy(qdev,
-					MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+					MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 				if (status)
 					goto exit;
 				ql_write32(qdev, MAC_ADDR_IDX, (offset++) | /* offset */
@@ -294,7 +294,7 @@ int ql_get_mac_addr_reg(struct ql_adapter *qdev, u32 type, u16 index,
 					   MAC_ADDR_ADR | MAC_ADDR_RS | type); /* type */
 				status =
 				    ql_wait_reg_rdy(qdev, MAC_ADDR_IDX,
-						    MAC_ADDR_MR, MAC_ADDR_E);
+						    MAC_ADDR_MR, 0);
 				if (status)
 					goto exit;
 				*value++ = ql_read32(qdev, MAC_ADDR_DATA);
@@ -344,7 +344,7 @@ static int ql_set_mac_addr_reg(struct ql_adapter *qdev, u8 *addr, u32 type,
 
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 			if (status)
 				goto exit;
 			ql_write32(qdev, MAC_ADDR_IDX, (offset++) | /* offset */
@@ -353,7 +353,7 @@ static int ql_set_mac_addr_reg(struct ql_adapter *qdev, u8 *addr, u32 type,
 			ql_write32(qdev, MAC_ADDR_DATA, lower);
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 			if (status)
 				goto exit;
 			ql_write32(qdev, MAC_ADDR_IDX, (offset++) | /* offset */
@@ -362,7 +362,7 @@ static int ql_set_mac_addr_reg(struct ql_adapter *qdev, u8 *addr, u32 type,
 			ql_write32(qdev, MAC_ADDR_DATA, upper);
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 			if (status)
 				goto exit;
 			ql_write32(qdev, MAC_ADDR_IDX, (offset) |	/* offset */
@@ -400,7 +400,7 @@ static int ql_set_mac_addr_reg(struct ql_adapter *qdev, u8 *addr, u32 type,
 
 			status =
 			    ql_wait_reg_rdy(qdev,
-				MAC_ADDR_IDX, MAC_ADDR_MW, MAC_ADDR_E);
+				MAC_ADDR_IDX, MAC_ADDR_MW, 0);
 			if (status)
 				goto exit;
 			ql_write32(qdev, MAC_ADDR_IDX, offset |	/* offset */
@@ -431,13 +431,13 @@ int ql_get_routing_reg(struct ql_adapter *qdev, u32 index, u32 *value)
 	if (status)
 		goto exit;
 
-	status = ql_wait_reg_rdy(qdev, RT_IDX, RT_IDX_MW, RT_IDX_E);
+	status = ql_wait_reg_rdy(qdev, RT_IDX, RT_IDX_MW, 0);
 	if (status)
 		goto exit;
 
 	ql_write32(qdev, RT_IDX,
 		   RT_IDX_TYPE_NICQ | RT_IDX_RS | (index << RT_IDX_IDX_SHIFT));
-	status = ql_wait_reg_rdy(qdev, RT_IDX, RT_IDX_MR, RT_IDX_E);
+	status = ql_wait_reg_rdy(qdev, RT_IDX, RT_IDX_MR, 0);
 	if (status)
 		goto exit;
 	*value = ql_read32(qdev, RT_DATA);
-- 
1.6.0

Re: [PATCH 5/5] qlge: bugfix: Fix register access error checking.

[David Miller]

From: Ron Mercer <ron.mercer@qlogic.com>
Date: Tue, 30 Dec 2008 13:12:25 -0800

> Some indexed registers do not have error bits. In these cases a
> value of zero should be used for error checking.
> 
> Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>

Also applied, thanks Ron.

[PATCH 0/5] qlge: Fixes for powerpc platform.

[Ron Mercer]

The following 5 patches fix bugs found during bringup on PPC platform.

[PATCH 2/5] qlge: bugfix: Add missing pci_unmap_page call in receive path.

[Ron Mercer]

Signed-off-by: Ron Mercer <ron.mercer@qlogic.com>
---
 drivers/net/qlge/qlge_main.c |    5 +++++
 1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/drivers/net/qlge/qlge_main.c b/drivers/net/qlge/qlge_main.c
index 56c7531..92dc3f7 100755
--- a/drivers/net/qlge/qlge_main.c
+++ b/drivers/net/qlge/qlge_main.c
@@ -1308,6 +1308,11 @@ static struct sk_buff *ql_build_rx_skb(struct ql_adapter *qdev,
 					"No skb available, drop the packet.\n");
 				return NULL;
 			}
+			pci_unmap_page(qdev->pdev,
+				       pci_unmap_addr(lbq_desc,
+						      mapaddr),
+				       pci_unmap_len(lbq_desc, maplen),
+				       PCI_DMA_FROMDEVICE);
 			skb_reserve(skb, NET_IP_ALIGN);
 			QPRINTK(qdev, RX_STATUS, DEBUG,
 				"%d bytes of headers and data in large. Chain page to new skb and pull tail.\n", length);
-- 
1.6.0