PATCH: ehci-q.c

Hi all,

    I apologize for my previous unreadable message... sorry.

    I'm workig on a porting on linux of a device driver for a double-side 
scanner.

    This device is equipped by a Cypress FX2 usb2.0 controller that, by the 
OEM, is configured
for bulk I/O with EndPoint "MaxPacketSize=1024", non usb2.0-compliant but 
supported by hardware and software device drivers on MS-Windows systems.

This problem is known (cfr the J. C. Andrus - 
http://lkml.org/lkml/2007/4/13/199) but is still present also in the recent 
kernel release: When the device send packets which were 1024 bytes in size, 
the urb came back with a status set to -EOVERFLOW.

    Instead of the solution proposed by J. C. Andrus, that is implemented 
for backward compatibility with a special urb flag that implies a more 
complex coding of bulk transfers, I made a simple "Endpoint Descriptor" 
driven solution, using the same logic alwais used for "INTERRUPT" EndPoint, 
that is still backward-compatible, but supports also new devices 
non-usb2.0-compliants.

    This solution is also "transparent" to current usb libraries and no 
special urb-flags are required, permitting the use of the high-level 
functions of current usblibs.

    In all tests I made, all USB devices working with Bulk transfer I 
connected (External HD&DVD, PenDrives, TV-Tuner and SkypePhone) are still 
working with their alredy installed drivers.

    It's possible to include this patch (or an analogue but working 
solution) in next releases of usb kernel module ?


All comments will be appreciated !

Thanks and ciao,
  Sergio

_____________________________________________________________________________________________

--- ehci-q.c_old 2007-07-09 01:32:17.000000000 +0200
+++ ehci-q.c 2007-11-23 02:16:14.000000000 +0100
@@ -20,6 +20,14 @@

/*-------------------------------------------------------------------------*/

+/* NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE NOTE 
NOTE
+ * Patched by S.Arcangeli to let it work with Cypress FX2 chipset that 
allows
+ * non 2.0-compliant Bulk transfers. The queue always uses the 
MaxPacketSize
+ * value declared by the EndPoint Descriptor instead of the 512 Fixed 
value.
+ * By this way the queue is well working also with these non standard 
devices.
+ * See down rows 459-460, 764-765
+ */
+
 /*
  * EHCI hardware queue manipulation ... the core.  QH/QTD manipulation.
  *
@@ -448,6 +456,8 @@
 #define hb_mult(wMaxPacketSize) (1 + (((wMaxPacketSize) >> 11) & 0x03))
 // ... and packet size, for any kind of endpoint descriptor
 #define max_packet(wMaxPacketSize) ((wMaxPacketSize) & 0x07ff)
+// ... but bulk endpoints can work only with packetsize of 512 or 1024 
bytes
+#define max_bulkpacket(wMaxPacketSize) (max_packet(wMaxPacketSize) > 512 ? 
1024 : 512)

 /*
  * reverse of qh_urb_transaction:  free a list of TDs.
@@ -751,7 +761,8 @@
                         info2 |= (EHCI_TUNE_MULT_HS << 30);
                 } else if (type == PIPE_BULK) {
                         info1 |= (EHCI_TUNE_RL_HS << 28);
-                        info1 |= 512 << 16;     /* usb2 fixed maxpacket */
+                        //info1 |= 512 << 16;   /* Original source: 
strictly usb2.0 compliant 512 bytes fixed maxpacket */
+                        info1 |= max_bulkpacket(maxp) << 16;  /* Uses the 
maxpacket from EP Descriptor, can be only 1024 or 512*/
                         info2 |= (EHCI_TUNE_MULT_HS << 30);
                 } else {                /* PIPE_INTERRUPT */
                         info1 |= max_packet (maxp) << 16;
_____________________________________________________________________________________________ 


-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel