[PATCH] bgmac: ratelimit warning messages

On my test environment the troughput of a file transfer drops from
4.4Mbps to 116Kbps due the number of repeated warning messages.
Adding printk_ratelimit() solves the issue without removing the
warning message.

Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
---
 drivers/net/ethernet/broadcom/bgmac.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bgmac.c b/drivers/net/ethernet/broadcom/bgmac.c
index 676ffe0..d1a259b 100644
--- a/drivers/net/ethernet/broadcom/bgmac.c
+++ b/drivers/net/ethernet/broadcom/bgmac.c

@@ -302,7 +302,7 @@ static int bgmac_dma_rx_skb_for_slot(struct bgmac *bgmac,
 	slot->skb = skb;
 	slot->dma_addr = dma_addr;
 
-	if (slot->dma_addr & 0xC0000000)
+	if ((slot->dma_addr & 0xC0000000) && printk_ratelimit())
 		bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");
 
 	return 0;

@@ -505,7 +505,7 @@ static int bgmac_dma_alloc(struct bgmac *bgmac)
 				  ring->mmio_base);
 			goto err_dma_free;
 		}
-		if (ring->dma_base & 0xC0000000)
+		if ((ring->dma_base & 0xC0000000) && printk_ratelimit())
 			bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");
 
 		ring->unaligned = bgmac_dma_unaligned(bgmac, ring,

@@ -536,7 +536,7 @@ static int bgmac_dma_alloc(struct bgmac *bgmac)
 			err = -ENOMEM;
 			goto err_dma_free;
 		}
-		if (ring->dma_base & 0xC0000000)
+		if ((ring->dma_base & 0xC0000000) && printk_ratelimit())
 			bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");
 
 		ring->unaligned = bgmac_dma_unaligned(bgmac, ring,

-- 
2.1.0

On Fri, 2015-03-06 at 19:17 +0100, Peter Senna Tschudin wrote:

On my test environment the troughput of a file transfer drops from
4.4Mbps to 116Kbps due the number of repeated warning messages.
Adding printk_ratelimit() solves the issue without removing the
warning message.

What value is there for this message being emitted more than once?

Maybe better as a static function

static void check_translate_address(struct bgmac *bgmac, dma_addr_t addr)
{
	static bool printed;

	if (!printed && (addr & (dma_addr_t)0xC0000000)))
		printed = true;
		bgmac_warn(bgmac, etc...);
	}
}

[...]

	check_translate_address(bgmac, slot->dma_addr);

On 6 March 2015 at 19:17, Peter Senna Tschudin [off-list ref] wrote:

Nack. Please just drop this code.

From what I understand network works for you. I've also checked

Broadcom's open source code, and on SoC we can always safely write
whole low 32b.

On Sat, Mar 7, 2015 at 12:27 AM, Rafał Miłecki [off-list ref] wrote:

On 6 March 2015 at 19:17, Peter Senna Tschudin [off-list ref] wrote:

quoted

On my test environment the troughput of a file transfer drops from
4.4Mbps to 116Kbps due the number of repeated warning messages.
Adding printk_ratelimit() solves the issue without removing the
warning message.

Signed-off-by: Peter Senna Tschudin <peter.senna@gmail.com>
---
 drivers/net/ethernet/broadcom/bgmac.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bgmac.c b/drivers/net/ethernet/broadcom/bgmac.c
index 676ffe0..d1a259b 100644
--- a/drivers/net/ethernet/broadcom/bgmac.c
+++ b/drivers/net/ethernet/broadcom/bgmac.c

@@ -302,7 +302,7 @@ static int bgmac_dma_rx_skb_for_slot(struct bgmac *bgmac,
        slot->skb = skb;
        slot->dma_addr = dma_addr;

-       if (slot->dma_addr & 0xC0000000)
+       if ((slot->dma_addr & 0xC0000000) && printk_ratelimit())
                bgmac_warn(bgmac, "DMA address using 0xC0000000 bit(s), it may need translation trick\n");

        return 0;

Nack. Please just drop this code.

From what I understand network works for you. I've also checked
Broadcom's open source code, and on SoC we can always safely write
whole low 32b.

Yes, it works for me, even after:

-CONFIG_PAGE_OFFSET=0xC0000000
+CONFIG_PAGE_OFFSET=0x40000000
...
+CONFIG_VMSPLIT_1G=y
+# CONFIG_VMSPLIT_3G is not set

The page offset change was made by make menuconfig as a consequence of
CONFIG_VMSPLIT_1G. After the Kconfig changes the throughput is almost
40 times worst, but commenting the warnings fixes the issue. I'll send
V2 removing the messages completely.


-- 
Peter