[bug report] wcn36xx: check dma_mapping_error()

[bug report] wcn36xx: check dma_mapping_error()

[Dan Carpenter]

Hello Rob Clark,

The patch 4165cf7ba52a: "wcn36xx: check dma_mapping_error()" from Jul
3, 2017, leads to the following Smatch static checker warning:

    drivers/net/wireless/ath/wcn36xx/dxe.c:305 wcn36xx_dxe_fill_skb()
    warn: 'dxe->dst_addr_l' is not a DMA mapping error

    drivers/net/wireless/ath/wcn36xx/dxe.c:805 wcn36xx_dxe_tx_frame()
    warn: 'desc_skb->src_addr_l' is not a DMA mapping error

drivers/net/wireless/ath/wcn36xx/dxe.c
    290 static int wcn36xx_dxe_fill_skb(struct device *dev,
    291                                 struct wcn36xx_dxe_ctl *ctl,
    292                                 gfp_t gfp)
    293 {
    294         struct wcn36xx_dxe_desc *dxe = ctl->desc;
    295         struct sk_buff *skb;
    296 
    297         skb = alloc_skb(WCN36XX_PKT_SIZE, gfp);
    298         if (skb == NULL)
    299                 return -ENOMEM;
    300 
    301         dxe->dst_addr_l = dma_map_single(dev,
    302                                          skb_tail_pointer(skb),
    303                                          WCN36XX_PKT_SIZE,
    304                                          DMA_FROM_DEVICE);
--> 305         if (dma_mapping_error(dev, dxe->dst_addr_l)) {

I'm so surprised that adding this check fixed a bug.  There is an
addr_l and addr_h (presumably for the low and high bits of the address?)
The ->dst_addr_h is never used.

A 32bit dst_addr_l can't be equal to DMA_MAPPING_ERROR unless it's a 32
bit system-ish.

    306                 dev_err(dev, "unable to map skb\n");
    307                 kfree_skb(skb);
    308                 return -ENOMEM;
    309         }
    310         ctl->skb = skb;
    311 
    312         return 0;
    313 }

regards,
dan carpenter

Re: [bug report] wcn36xx: check dma_mapping_error()

[Rob Clark]

On Thu, Aug 25, 2022 at 9:10 AM Dan Carpenter <dan.carpenter@oracle.com> wrote:
>
> Hello Rob Clark,
>
> The patch 4165cf7ba52a: "wcn36xx: check dma_mapping_error()" from Jul
> 3, 2017, leads to the following Smatch static checker warning:
>
>     drivers/net/wireless/ath/wcn36xx/dxe.c:305 wcn36xx_dxe_fill_skb()
>     warn: 'dxe->dst_addr_l' is not a DMA mapping error
>
>     drivers/net/wireless/ath/wcn36xx/dxe.c:805 wcn36xx_dxe_tx_frame()
>     warn: 'desc_skb->src_addr_l' is not a DMA mapping error
>
> drivers/net/wireless/ath/wcn36xx/dxe.c
>     290 static int wcn36xx_dxe_fill_skb(struct device *dev,
>     291                                 struct wcn36xx_dxe_ctl *ctl,
>     292                                 gfp_t gfp)
>     293 {
>     294         struct wcn36xx_dxe_desc *dxe = ctl->desc;
>     295         struct sk_buff *skb;
>     296
>     297         skb = alloc_skb(WCN36XX_PKT_SIZE, gfp);
>     298         if (skb == NULL)
>     299                 return -ENOMEM;
>     300
>     301         dxe->dst_addr_l = dma_map_single(dev,
>     302                                          skb_tail_pointer(skb),
>     303                                          WCN36XX_PKT_SIZE,
>     304                                          DMA_FROM_DEVICE);
> --> 305         if (dma_mapping_error(dev, dxe->dst_addr_l)) {
>
> I'm so surprised that adding this check fixed a bug.  There is an
> addr_l and addr_h (presumably for the low and high bits of the address?)
> The ->dst_addr_h is never used.
>
> A 32bit dst_addr_l can't be equal to DMA_MAPPING_ERROR unless it's a 32
> bit system-ish.

judging by the date, it was probably on a 32b system ;-)

that said, it doesn't seem like that code even supports systems w/ 64b dma addr

BR,
-R

>
>     306                 dev_err(dev, "unable to map skb\n");
>     307                 kfree_skb(skb);
>     308                 return -ENOMEM;
>     309         }
>     310         ctl->skb = skb;
>     311
>     312         return 0;
>     313 }
>
> regards,
> dan carpenter