Commit 66600fac7a98 ("net: stmmac: TSO: Fix unbalanced DMA map/unmap
for non-paged SKB data") moved the assignment of tx_skbuff_dma[]'s
members to be later in stmmac_tso_xmit().

The buf (dma cookie) and len stored in this structure are passed to
dma_unmap_single() by stmmac_tx_clean(). The DMA API requires that
the dma cookie passed to dma_unmap_single() is the same as the value
returned from dma_map_single(). However, by moving the assignment
later, this is not the case when priv->dma_cap.addr64 > 32 as "des"
is offset by proto_hdr_len.

This causes problems such as:

  dwc-eth-dwmac 2490000.ethernet eth0: Tx DMA map failed

and with DMA_API_DEBUG enabled:

  DMA-API: dwc-eth-dwmac 2490000.ethernet: device driver tries to +free DMA memory it has not allocated [device address=0x000000ffffcf65c0] [size=66 bytes]

Fix this by maintaining "des" as the original DMA cookie, and use
tso_des to pass the offset DMA cookie to stmmac_tso_allocator().

Full details of the crashes can be found at:
https://lore.kernel.org/all/d8112193-0386-4e14-b516-37c2d838171a@nvidia.com/
https://lore.kernel.org/all/klkzp5yn5kq5efgtrow6wbvnc46bcqfxs65nz3qy77ujr5turc@bwwhelz2l4dw/

Reported-by: Jon Hunter <jonathanh@nvidia.com>
Reported-by: Thierry Reding <thierry.reding@gmail.com>
Fixes: 66600fac7a98 ("net: stmmac: TSO: Fix unbalanced DMA map/unmap for non-paged SKB data")
Tested-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
---
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index 9b262cdad60b..c81ea8cdfe6e 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -4192,8 +4192,8 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
 	struct stmmac_txq_stats *txq_stats;
 	struct stmmac_tx_queue *tx_q;
 	u32 pay_len, mss, queue;
+	dma_addr_t tso_des, des;
 	u8 proto_hdr_len, hdr;
-	dma_addr_t des;
 	bool set_ic;
 	int i;
 
@@ -4289,14 +4289,15 @@ static netdev_tx_t stmmac_tso_xmit(struct sk_buff *skb, struct net_device *dev)
 
 		/* If needed take extra descriptors to fill the remaining payload */
 		tmp_pay_len = pay_len - TSO_MAX_BUFF_SIZE;
+		tso_des = des;
 	} else {
 		stmmac_set_desc_addr(priv, first, des);
 		tmp_pay_len = pay_len;
-		des += proto_hdr_len;
+		tso_des = des + proto_hdr_len;
 		pay_len = 0;
 	}
 
-	stmmac_tso_allocator(priv, des, tmp_pay_len, (nfrags == 0), queue);
+	stmmac_tso_allocator(priv, tso_des, tmp_pay_len, (nfrags == 0), queue);
 
 	/* In case two or more DMA transmit descriptors are allocated for this
 	 * non-paged SKB data, the DMA buffer address should be saved to
-- 
2.30.2

On Fri, 06 Dec 2024 12:40:11 +0000, "Russell King (Oracle)" <rmk+kernel@armlinux.org.uk> wrote:

> Commit 66600fac7a98 ("net: stmmac: TSO: Fix unbalanced DMA map/unmap
> for non-paged SKB data") moved the assignment of tx_skbuff_dma[]'s
> members to be later in stmmac_tso_xmit().
> 
> The buf (dma cookie) and len stored in this structure are passed to
> dma_unmap_single() by stmmac_tx_clean(). The DMA API requires that
> the dma cookie passed to dma_unmap_single() is the same as the value
> returned from dma_map_single(). However, by moving the assignment
> later, this is not the case when priv->dma_cap.addr64 > 32 as "des"
> is offset by proto_hdr_len.
> 
> This causes problems such as:
> 
>   dwc-eth-dwmac 2490000.ethernet eth0: Tx DMA map failed
> 
> and with DMA_API_DEBUG enabled:
> 
>   DMA-API: dwc-eth-dwmac 2490000.ethernet: device driver tries to +free DMA memory it has not allocated [device address=0x000000ffffcf65c0] [size=66 bytes]
> 
> Fix this by maintaining "des" as the original DMA cookie, and use
> tso_des to pass the offset DMA cookie to stmmac_tso_allocator().
> 
> Full details of the crashes can be found at:
> https://lore.kernel.org/all/d8112193-0386-4e14-b516-37c2d838171a@nvidia.com/
> https://lore.kernel.org/all/klkzp5yn5kq5efgtrow6wbvnc46bcqfxs65nz3qy77ujr5turc@bwwhelz2l4dw/
> 
> Reported-by: Jon Hunter <jonathanh@nvidia.com>
> Reported-by: Thierry Reding <thierry.reding@gmail.com>
> Fixes: 66600fac7a98 ("net: stmmac: TSO: Fix unbalanced DMA map/unmap for non-paged SKB data")

Much appreciated for this fix.

Reviewed-by: Furong Xu <0x1207@gmail.com>