ks7010 endianness question

ks7010 endianness question

[Tobin C. Harding]

Hi Wolfram,

May I please ask you with an ks7010 driver endianness question?

Comments on the hostif_hdr data structure (ks_hostif.h) state that the
target uses little endian byte order.

/*
 * HOST-MAC I/F data structure
 * Byte alignmet Little Endian
 */

struct hostif_hdr {
	u16 size;
	u16 event;
} __packed;

On the rx data path this header is unpacked using get_WORD()

void hostif_receive(struct ks_wlan_private *priv, unsigned char *p,
		    unsigned int size)
{
	DPRINTK(4, "\n");

	devio_rec_ind(priv, p, size);

	priv->rxp = p;
	priv->rx_size = size;

	if (get_WORD(priv) == priv->rx_size) {	/* length check !! */
		hostif_event_check(priv);	/* event check */
	}
}

get_WORD() inverts the byte order

static inline u16 get_WORD(struct ks_wlan_private *priv)
{
	u16 data;

	data = (get_BYTE(priv) & 0xff);
	data |= ((get_BYTE(priv) << 8) & 0xff00);
	return data;
}

Am I missing something? It seems that this code will only work if the
host and the target have differing endianness. It seems unlikely that
the driver was tested solely on a big-endian machine, is the comment
wrong - is the target actually big-endian?

Off topic, I watched your 2014 Fosdem talk on adding device support to
the kernel without adding code. It was very educational.

thanks for your time,
Tobin.

Re: ks7010 endianness question

[Tobin C. Harding]

On Wed, Apr 19, 2017 at 01:34:46PM +1000, Tobin C. Harding wrote:
> Hi Wolfram,
> 
> May I please ask you with an ks7010 driver endianness question?
> 
> Comments on the hostif_hdr data structure (ks_hostif.h) state that the
> target uses little endian byte order.
> 
> /*
>  * HOST-MAC I/F data structure
>  * Byte alignmet Little Endian
>  */
> 
> struct hostif_hdr {
> 	u16 size;
> 	u16 event;
> } __packed;
> 
> On the rx data path this header is unpacked using get_WORD()
> 
> void hostif_receive(struct ks_wlan_private *priv, unsigned char *p,
> 		    unsigned int size)
> {
> 	DPRINTK(4, "\n");
> 
> 	devio_rec_ind(priv, p, size);
> 
> 	priv->rxp = p;
> 	priv->rx_size = size;
> 
> 	if (get_WORD(priv) == priv->rx_size) {	/* length check !! */
> 		hostif_event_check(priv);	/* event check */
> 	}
> }
> 
> get_WORD() inverts the byte order
> 
> static inline u16 get_WORD(struct ks_wlan_private *priv)
> {
> 	u16 data;
> 
> 	data = (get_BYTE(priv) & 0xff);
> 	data |= ((get_BYTE(priv) << 8) & 0xff00);
> 	return data;
> }
> 
> Am I missing something? It seems that this code will only work if the
> host and the target have differing endianness. It seems unlikely that
> the driver was tested solely on a big-endian machine, is the comment
> wrong - is the target actually big-endian?

Further investigation suggests that the target is actually using
network byte order (implying that the comment is wrong).

static void hostif_data_indication(struct ks_wlan_private *priv)
{
        ...
        
	auth_type = get_WORD(priv);	/* AuthType */
	get_WORD(priv);	                /* Reserve Area */

	eth_hdr = (struct ether_hdr *)(priv->rxp);
	eth_proto = ntohs(eth_hdr->h_proto);

        ...


I think we can call this question resolved. Remove the comment and
change the hostif_hdr description to

struct hostif_hdr {
	__be16 size;
 	__be16 event;
} __packed;

Are you happy with this?

thanks,
Tobin.

Re: ks7010 endianness question

[Wolfram Sang]

[-- Attachment #1: Type: text/plain, Size: 245 bytes --]


> I think we can call this question resolved. Remove the comment and
> change the hostif_hdr description to
> 
> struct hostif_hdr {
> 	__be16 size;
>  	__be16 event;
> } __packed;
> 
> Are you happy with this?

Yes, I trust you.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]