Date:   Thu, 23 Mar 2023 18:22:23 +0900
From:   Vincent MAILHOL <mailhol.vincent@...adoo.fr>
To:     Peter Hong <peter_hong@...tek.com.tw>
Cc:     wg@...ndegger.com, mkl@...gutronix.de,
        michal.swiatkowski@...ux.intel.com, Steen.Hegelund@...rochip.com,
        davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
        pabeni@...hat.com, frank.jungclaus@....eu,
        linux-kernel@...r.kernel.org, linux-can@...r.kernel.org,
        netdev@...r.kernel.org, hpeter+linux_kernel@...il.com
Subject: Re: [PATCH V2] can: usb: f81604: add Fintek F81604 support

On Thu. 23 Mar 2023 at 14:54, Vincent MAILHOL
<mailhol.vincent@...adoo.fr> wrote:
> Le jeu. 23 mars 2023 à 14:14, Peter Hong <peter_hong@...tek.com.tw> a écrit :
> >
> > Hi Vincent,
> >
> > Vincent MAILHOL 於 2023/3/21 下午 11:50 寫道:
> > >> +static netdev_tx_t f81604_start_xmit(struct sk_buff *skb,
> > >> +                                    struct net_device *netdev)
> > >> +{
> > >> +       struct can_frame *cf = (struct can_frame *)skb->data;
> > >> +       struct f81604_port_priv *priv = netdev_priv(netdev);
> > >> +       struct net_device_stats *stats = &netdev->stats;
> > >> +       int status;
> > >> +       u8 *ptr;
> > >> +       u32 id;
> > >> +
> > >> +       if (can_dropped_invalid_skb(netdev, skb))
> > >> +               return NETDEV_TX_OK;
> > >> +
> > >> +       netif_stop_queue(netdev);
> > >> +
> > >> +       ptr = priv->bulk_write_buffer;
> > >> +       memset(ptr, 0, F81604_DATA_SIZE);
> > >> +
> > >> +       ptr[0] = F81604_CMD_DATA;
> > >> +       ptr[1] = min_t(u8, cf->can_dlc & 0xf, 8);
> > >> +
> > >> +       if (cf->can_id & CAN_EFF_FLAG) {
> > >> +               id = (cf->can_id & CAN_ERR_MASK) << 3;
> > >> +               ptr[1] |= F81604_EFF_BIT;
> > >> +               ptr[2] = (id >> 24) & 0xff;
> > >> +               ptr[3] = (id >> 16) & 0xff;
> > >> +               ptr[4] = (id >> 8) & 0xff;
> > >> +               ptr[5] = (id >> 0) & 0xff;
> > >> +               memcpy(&ptr[6], cf->data, ptr[1]);
> > > Rather than manipulating an opaque u8 array, please declare a
> > > structure with explicit names.
> >
> > I had try to declare a struct like below and refactoring code :
> >
> > struct f81604_bulk_data {
> >      u8 cmd;
> >      u8 dlc;
> >
> >      union {
> >          struct {
> >              u8 id1, id2;
> >              u8 data[CAN_MAX_DLEN];
> >          } sff;
> >
> >          struct {
> >              u8 id1, id2, id3, id4;
> >              u8 data[CAN_MAX_DLEN];
> >          } eff;
> >      };
> > } __attribute__((packed));
> >
> > This struct can used in TX/RX bulk in/out. Is it ok?
>
> That's nearly it. It is better to declare the struct sff and eff
> separately. Also, do not split the id in bytes. Declare it as a little
> endian using the __le16 and __le32 types.
>
> Something like this (I let you adjust):
>
>   struct f81604_sff {
>           __le16 id:
>           u8 data[CAN_MAX_DLEN];
>   } __attribute__((packed));
>
>   struct f81604_eff {
>           __le32 id;
>           u8 data[CAN_MAX_DLEN];
>   } __attribute__((packed));
>
>   struct f81604_bulk_data {
>           u8 cmd;
>           u8 dlc;
>
>           union {
>                   struct f81604_sff sff;
>                   struct f81604_eff eff;
>            };
>   } __attribute__((packed));
>
> The __le16 field should be manipulated using cpu_to_leXX() and
> leXX_to_cpu() if the field is aligned, if not, it should be
> manipulated using {get|set}_unaligned_leXX() (where XX represents the
> size in bits).
>
> Also, f81604_bulk_data->dlc is not only a DLC but also carries the
> F81604_EFF_BIT flag, right? At least, add documentation to the
> structure to clarify this point.
>
> > > +static int f81604_prepare_urbs(struct net_device *netdev)
> > > +{
> > > +       static const u8 bulk_in_addr[F81604_MAX_DEV] = { 0x82, 0x84 };
> > > +       static const u8 bulk_out_addr[F81604_MAX_DEV] = { 0x01, 0x03 };
> > > +       static const u8 int_in_addr[F81604_MAX_DEV] = { 0x81, 0x83 };
> > > +       struct f81604_port_priv *priv = netdev_priv(netdev);
> > > +       int id = netdev->dev_id;
> > > +       int i;
> > > +
> > > +       /* initialize to NULL for error recovery */
> > > +       for (i = 0; i < F81604_MAX_RX_URBS; ++i)
> > > +               priv->read_urb[i] = NULL;
> > > priv was allocated with devm_kzalloc() so it should already be zeroed,
> > > right? What is the purpose of this loop?
> >
> > This operation due to following condition:
> >      f81604_open() -> f81604_close() -> f81604_open() failed.
> >
> > We had used  devm_kzalloc() in f81604_probe(), so first f81604_open() all
> > pointers are NULL. But after f81604_close() then f81604_open() second
> > times, the URB pointers are not NULLed, it'll makes error on 2nd
> > f81604_open()
> > with fail.
>
> Makes sense, thanks for the clarification.
>
> Then, please replace your loop by a memset(priv->read_urb, 0,
> sizeof(priv->read_urb).

Actually, your code never accesses the zeroed memory. The next lines are:

  for (i = 0; i < F81604_MAX_RX_URBS; ++i) {
          priv->read_urb[i] = usb_alloc_urb(0, GFP_KERNEL);

If priv->read_urb[i] is never read before being initialized, no need to zero it.

> > >> +/* Called by the usb core when driver is unloaded or device is removed */
> > >> +static void f81604_disconnect(struct usb_interface *intf)
> > >> +{
> > >> +       struct f81604_priv *priv = usb_get_intfdata(intf);
> > >> +       int i;
> > >> +
> > >> +       for (i = 0; i < F81604_MAX_DEV; ++i) {
> > >> +               if (!priv->netdev[i])
> > >> +                       continue;
> > >> +
> > >> +               unregister_netdev(priv->netdev[i]);
> > >> +               free_candev(priv->netdev[i]);
> > >> +       }
> > >   i> +}
> >
> > Is typo here?
>
> Yes, please ignore.
>
>
> Yours sincerely,
> Vincent Mailhol

Powered by blists - more mailing lists