Re: [PATCH net] ch9200: return error on failed register writes in ch9200_bind()

[netdev-bot+sashiko@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential
issue(s) to consider:

Critical: 0 · High: 1 · Medium: 0 · Low: 0

- [High] Ineffective fix: the six newly added `if (retval < 0) return
  retval;` checks in ch9200_bind() (drivers/net/usb/ch9200.c) cannot…

--

ch9200: return error on failed register writes in ch9200_bind()

The patch adds error checks after the six control_write() calls in
ch9200_bind() so that a failed vendor register write aborts the bind
instead of being ignored.  It carries a Fixes tag for the original driver
commit 4a476bd6d1d9.

> diff --git a/drivers/net/usb/ch9200.c b/drivers/net/usb/ch9200.c
> index a206ffa76f1b9..ab3cd3902ed72 100644
> --- a/drivers/net/usb/ch9200.c
> +++ b/drivers/net/usb/ch9200.c
> @@ -360,32 +360,44 @@ static int ch9200_bind(struct usbnet *dev, struct usb_interface *intf)
>  	data[1] = 0x0F;
>  	retval = control_write(dev, REQUEST_WRITE, 0, MAC_REG_THRESHOLD, data,
>  			       0x02, CONTROL_TIMEOUT_MS);
> +	if (retval < 0)
> +		return retval;

[Severity: High]

Can these new checks ever observe a failed register write?  control_write()
in the same file discards the USB transfer status on its normal path:

drivers/net/usb/ch9200.c:control_write() {
	...
	err = usb_control_msg(dev->udev,
			      usb_sndctrlpipe(dev->udev, 0),
			      request, request_type, value, index, buf, size,
			      timeout);
	if (err >= 0 && err < size)
		err = -EINVAL;
	kfree(buf);

	return 0;

err_out:
	return err;
}

The only return err; is at err_out:, which is reached solely from the
kmemdup() -ENOMEM branch.  All six patched call sites pass a non-NULL
2-byte data buffer, so the only negative value the new checks can see is
-ENOMEM from that allocation.

A STALL of the vendor OUT request (-EPIPE), a timeout (-ETIMEDOUT), a
disconnect during bind (-ENODEV/-ESHUTDOWN), a protocol error (-EPROTO),
or a short control transfer (normalized to -EINVAL just above) are all
turned into 0, so ch9200_bind() still falls through to
get_mac_address()/eth_hw_addr_set() and reports success with
MAC_REG_THRESHOLD, MAC_REG_FIFO_DEPTH, MAC_REG_PAUSE,
MAC_REG_FLOW_CONTROL, register 254 and MAC_REG_CTRL (which arms TX/RX)
left unprogrammed.

The sibling helper control_read() already returns err on its normal path,
and ch9200_mdio_read() checks it with if (ret < 0), so the convention
control_write() is expected to follow already exists in this file.

Would it make more sense to change control_write() to return err after
kfree(buf) instead of (or in addition to) adding the caller-side checks?
As it stands the commit message claim that this makes the driver "bail out
of ch9200_bind() on failed writes" does not seem to hold for actual
transfer failures.

> 
>  	data[0] = 0xA0;
>  	data[1] = 0x90;
>  	retval = control_write(dev, REQUEST_WRITE, 0, MAC_REG_FIFO_DEPTH, data,
>  			       0x02, CONTROL_TIMEOUT_MS);
> +	if (retval < 0)
> +		return retval;

[ ... remaining four checks have the same question ... ]

>  	retval = get_mac_address(dev, addr);
>  	eth_hw_addr_set(dev->net, addr);

-- 
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20260914202503.52920-1-s.shtylyov%40auroraos.dev