[PATCH AUTOSEL 6.5 05/30] atl1c: Work around the DMA RX overflow issue

From: Sasha Levin
Date: Tue Nov 07 2023 - 07:11:37 EST


From: Sieng-Piaw Liew <liew.s.piaw@xxxxxxxxx>

[ Upstream commit 86565682e9053e5deb128193ea9e88531bbae9cf ]

This is based on alx driver commit 881d0327db37 ("net: alx: Work around
the DMA RX overflow issue").

The alx and atl1c drivers had RX overflow error which was why a custom
allocator was created to avoid certain addresses. The simpler workaround
then created for alx driver, but not for atl1c due to lack of tester.

Instead of using a custom allocator, check the allocated skb address and
use skb_reserve() to move away from problematic 0x...fc0 address.

Tested on AR8131 on Acer 4540.

Signed-off-by: Sieng-Piaw Liew <liew.s.piaw@xxxxxxxxx>
Link: https://lore.kernel.org/r/20230912010711.12036-1-liew.s.piaw@xxxxxxxxx
Signed-off-by: Paolo Abeni <pabeni@xxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
drivers/net/ethernet/atheros/atl1c/atl1c.h | 3 -
.../net/ethernet/atheros/atl1c/atl1c_main.c | 67 +++++--------------
2 files changed, 16 insertions(+), 54 deletions(-)

diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c.h b/drivers/net/ethernet/atheros/atl1c/atl1c.h
index 43d821fe7a542..63ba64dbb7310 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c.h
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c.h
@@ -504,15 +504,12 @@ struct atl1c_rrd_ring {
u16 next_to_use;
u16 next_to_clean;
struct napi_struct napi;
- struct page *rx_page;
- unsigned int rx_page_offset;
};

/* board specific private data structure */
struct atl1c_adapter {
struct net_device *netdev;
struct pci_dev *pdev;
- unsigned int rx_frag_size;
struct atl1c_hw hw;
struct atl1c_hw_stats hw_stats;
struct mii_if_info mii; /* MII interface info */
diff --git a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
index 940c5d1ff9cfc..74b78164cf74a 100644
--- a/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
+++ b/drivers/net/ethernet/atheros/atl1c/atl1c_main.c
@@ -483,15 +483,10 @@ static int atl1c_set_mac_addr(struct net_device *netdev, void *p)
static void atl1c_set_rxbufsize(struct atl1c_adapter *adapter,
struct net_device *dev)
{
- unsigned int head_size;
int mtu = dev->mtu;

adapter->rx_buffer_len = mtu > AT_RX_BUF_SIZE ?
roundup(mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN, 8) : AT_RX_BUF_SIZE;
-
- head_size = SKB_DATA_ALIGN(adapter->rx_buffer_len + NET_SKB_PAD + NET_IP_ALIGN) +
- SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
- adapter->rx_frag_size = roundup_pow_of_two(head_size);
}

static netdev_features_t atl1c_fix_features(struct net_device *netdev,
@@ -964,7 +959,6 @@ static void atl1c_init_ring_ptrs(struct atl1c_adapter *adapter)
static void atl1c_free_ring_resources(struct atl1c_adapter *adapter)
{
struct pci_dev *pdev = adapter->pdev;
- int i;

dma_free_coherent(&pdev->dev, adapter->ring_header.size,
adapter->ring_header.desc, adapter->ring_header.dma);
@@ -977,12 +971,6 @@ static void atl1c_free_ring_resources(struct atl1c_adapter *adapter)
kfree(adapter->tpd_ring[0].buffer_info);
adapter->tpd_ring[0].buffer_info = NULL;
}
- for (i = 0; i < adapter->rx_queue_count; ++i) {
- if (adapter->rrd_ring[i].rx_page) {
- put_page(adapter->rrd_ring[i].rx_page);
- adapter->rrd_ring[i].rx_page = NULL;
- }
- }
}

/**
@@ -1754,48 +1742,11 @@ static inline void atl1c_rx_checksum(struct atl1c_adapter *adapter,
skb_checksum_none_assert(skb);
}

-static struct sk_buff *atl1c_alloc_skb(struct atl1c_adapter *adapter,
- u32 queue, bool napi_mode)
-{
- struct atl1c_rrd_ring *rrd_ring = &adapter->rrd_ring[queue];
- struct sk_buff *skb;
- struct page *page;
-
- if (adapter->rx_frag_size > PAGE_SIZE) {
- if (likely(napi_mode))
- return napi_alloc_skb(&rrd_ring->napi,
- adapter->rx_buffer_len);
- else
- return netdev_alloc_skb_ip_align(adapter->netdev,
- adapter->rx_buffer_len);
- }
-
- page = rrd_ring->rx_page;
- if (!page) {
- page = alloc_page(GFP_ATOMIC);
- if (unlikely(!page))
- return NULL;
- rrd_ring->rx_page = page;
- rrd_ring->rx_page_offset = 0;
- }
-
- skb = build_skb(page_address(page) + rrd_ring->rx_page_offset,
- adapter->rx_frag_size);
- if (likely(skb)) {
- skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
- rrd_ring->rx_page_offset += adapter->rx_frag_size;
- if (rrd_ring->rx_page_offset >= PAGE_SIZE)
- rrd_ring->rx_page = NULL;
- else
- get_page(page);
- }
- return skb;
-}
-
static int atl1c_alloc_rx_buffer(struct atl1c_adapter *adapter, u32 queue,
bool napi_mode)
{
struct atl1c_rfd_ring *rfd_ring = &adapter->rfd_ring[queue];
+ struct atl1c_rrd_ring *rrd_ring = &adapter->rrd_ring[queue];
struct pci_dev *pdev = adapter->pdev;
struct atl1c_buffer *buffer_info, *next_info;
struct sk_buff *skb;
@@ -1814,13 +1765,27 @@ static int atl1c_alloc_rx_buffer(struct atl1c_adapter *adapter, u32 queue,
while (next_info->flags & ATL1C_BUFFER_FREE) {
rfd_desc = ATL1C_RFD_DESC(rfd_ring, rfd_next_to_use);

- skb = atl1c_alloc_skb(adapter, queue, napi_mode);
+ /* When DMA RX address is set to something like
+ * 0x....fc0, it will be very likely to cause DMA
+ * RFD overflow issue.
+ *
+ * To work around it, we apply rx skb with 64 bytes
+ * longer space, and offset the address whenever
+ * 0x....fc0 is detected.
+ */
+ if (likely(napi_mode))
+ skb = napi_alloc_skb(&rrd_ring->napi, adapter->rx_buffer_len + 64);
+ else
+ skb = netdev_alloc_skb(adapter->netdev, adapter->rx_buffer_len + 64);
if (unlikely(!skb)) {
if (netif_msg_rx_err(adapter))
dev_warn(&pdev->dev, "alloc rx buffer failed\n");
break;
}

+ if (((unsigned long)skb->data & 0xfff) == 0xfc0)
+ skb_reserve(skb, 64);
+
/*
* Make buffer alignment 2 beyond a 16 byte boundary
* this will result in a 16 byte aligned IP header after
--
2.42.0