On Tue, Oct 22, 2013 at 05:22:08PM +0200, David Herrmann wrote:

Hi

On Tue, Oct 22, 2013 at 8:08 AM, Florian Echtler [off-list ref] wrote:

quoted

Hello Dmitry,

thanks for your quick feedback, a few questions below:

On 21.10.2013 18:20, Dmitry Torokhov wrote:

quoted

On Sun, Oct 20, 2013 at 06:49:11PM +0200, Florian Echtler wrote:

quoted

+/* read 512 bytes from endpoint 0x86 -> get header + blobs */
+struct sur40_header {
+
+    uint16_t type;       /* always 0x0001 */
+    uint16_t count;      /* count of blobs (if 0: continue prev. packet) */
+
+    uint32_t packet_id;
+
+    uint32_t timestamp;  /* milliseconds (inc. by 16 or 17 each frame) */
+    uint32_t unknown;    /* "epoch?" always 02/03 00 00 00 */

Proper internal kernel types are u8, u16, u32. For user-facing APIs
__u8, __u16, and __u32 should be used. Also, since this is data coming
directly off the wire, you should be using __le16, __le32, etc, and then
do __leXX_to_cpu() conversion before using it in calculations.

OK, I'll switch to u32 throughout (also for the float, I'll explain in a
commment). However, I haven't found a single other touchscreen driver
which uses __le32, even though they all probably process raw wire data -
can you suggest an example?

These are probably all broken or the hardware guarantees
cpu-byte-order. Anyway, what you should do is use __le16/32 for your
types which represent data from the device. Then call le16_to_cpu() on
these values to convert it to host byte-order. Something like this:

struct sur40_raw_header {
  __le16 type;
  __le32 unused;
  __le8 count;

Not the last one please ;)

Thanks.

-- 
Dmitry