Re: [PATCH net 2/2] net: ethernet: mediatek: fix inconsistency of port number carried in TXD

[David Miller]

From: 
Date: Tue, 11 Apr 2017 15:12:34 +0800

> @@ -410,12 +410,18 @@ struct mtk_hw_stats {
>   struct u64_stats_sync   syncp;
>  };
>  
> -/* PDMA descriptor can point at 1-2 segments. This enum allows us to track 
> how
> - * memory was allocated so that it can be freed properly
> - */
>  enum mtk_tx_flags {
> + /* PDMA descriptor can point at 1-2 segments. This enum allows us to
> +  * track how memory was allocated so that it can be freed properly.
> +  */
>   MTK_TX_FLAGS_SINGLE0= 0x01,
>   MTK_TX_FLAGS_PAGE0  = 0x02,
> +
> + /* MTK_TX_FLAGS_FPORTx allows tracking which port the transmitted
> +  * SKB out instead of looking up through hardware TX descriptor.
> +  */
> + MTK_TX_FLAGS_FPORT1 = 0x03,
> + MTK_TX_FLAGS_FPORT0 = 0x04,
>  };

You're creating a confusing situation here and asking for bugs to be introduced
into the driver.

The existing two enumeration values are "bit masks", but these two new values
are bit numbers and must be used with BIT().

Make them consistent, and I would suggest to make all of them bit masks, thus:

MTK_TX_FLAGS_FPORT1 = 0x04,
MTK_TX_FLAGS_FPORT0 = 0x08,

And then remove the use of BIT() to these values.

Thanks.

[PATCH net 2/2] net: ethernet: mediatek: fix inconsistency of port number carried in TXD

[sean.wang]

From: Sean Wang 

Fix port inconsistency on TXD due to hardware BUG that would cause
different port number is carried on the same TXD between tx_map()
and tx_unmap() with the iperf test. It would cause confusing BQL
logic which leads to kernel panic when dual GMAC runs concurrently.

Signed-off-by: Sean Wang 
---
 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 14 +-
 drivers/net/ethernet/mediatek/mtk_eth_soc.h | 12 +---
 2 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.c 
b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
index 48ba617..502956c 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.c
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.c
@@ -648,6 +648,8 @@ static int mtk_tx_map(struct sk_buff *skb, struct 
net_device *dev,
 
WRITE_ONCE(itxd->txd1, mapped_addr);
itx_buf->flags |= MTK_TX_FLAGS_SINGLE0;
+   itx_buf->flags |= (!mac->id) ? BIT(MTK_TX_FLAGS_FPORT0) :
+ BIT(MTK_TX_FLAGS_FPORT1);
dma_unmap_addr_set(itx_buf, dma_addr0, mapped_addr);
dma_unmap_len_set(itx_buf, dma_len0, skb_headlen(skb));
 
@@ -689,6 +691,9 @@ static int mtk_tx_map(struct sk_buff *skb, struct 
net_device *dev,
memset(tx_buf, 0, sizeof(*tx_buf));
tx_buf->skb = (struct sk_buff *)MTK_DMA_DUMMY_DESC;
tx_buf->flags |= MTK_TX_FLAGS_PAGE0;
+   tx_buf->flags |= (!mac->id) ? BIT(MTK_TX_FLAGS_FPORT0) :
+BIT(MTK_TX_FLAGS_FPORT1);
+
dma_unmap_addr_set(tx_buf, dma_addr0, mapped_addr);
dma_unmap_len_set(tx_buf, dma_len0, frag_map_size);
frag_size -= frag_map_size;
@@ -1011,17 +1016,16 @@ static int mtk_poll_tx(struct mtk_eth *eth, int budget)
 
while ((cpu != dma) && budget) {
u32 next_cpu = desc->txd2;
-   int mac;
+   int mac = 0;
 
desc = mtk_qdma_phys_to_virt(ring, desc->txd2);
if ((desc->txd3 & TX_DMA_OWNER_CPU) == 0)
break;
 
-   mac = (desc->txd4 >> TX_DMA_FPORT_SHIFT) &
-  TX_DMA_FPORT_MASK;
-   mac--;
-
tx_buf = mtk_desc_to_tx_buf(ring, desc);
+   if (tx_buf->flags & BIT(MTK_TX_FLAGS_FPORT1))
+   mac = 1;
+
skb = tx_buf->skb;
if (!skb) {
condition = 1;
diff --git a/drivers/net/ethernet/mediatek/mtk_eth_soc.h 
b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
index 996024d..7327b23 100644
--- a/drivers/net/ethernet/mediatek/mtk_eth_soc.h
+++ b/drivers/net/ethernet/mediatek/mtk_eth_soc.h
@@ -410,12 +410,18 @@ struct mtk_hw_stats {
struct u64_stats_sync   syncp;
 };
 
-/* PDMA descriptor can point at 1-2 segments. This enum allows us to track how
- * memory was allocated so that it can be freed properly
- */
 enum mtk_tx_flags {
+   /* PDMA descriptor can point at 1-2 segments. This enum allows us to
+* track how memory was allocated so that it can be freed properly.
+*/
MTK_TX_FLAGS_SINGLE0= 0x01,
MTK_TX_FLAGS_PAGE0  = 0x02,
+
+   /* MTK_TX_FLAGS_FPORTx allows tracking which port the transmitted
+* SKB out instead of looking up through hardware TX descriptor.
+*/
+   MTK_TX_FLAGS_FPORT1 = 0x03,
+   MTK_TX_FLAGS_FPORT0 = 0x04,
 };
 
 /* This enum allows us to identify how the clock is defined on the array of the
-- 
1.9.1