[EMAIL PROTECTED] said:
> Hi all,
> sins I upgraded from 2.4.7 to 2.4.16 my self constructed USB device
> is very slow (~ 25% of the old speed) handled by the kernel. I also
> tested this with 2.4.18 and found no difference.
> My computers are (a) a laptop with some Intel chipset (usb-uhci) and
> (b) a desktop with some VIA chipsed (usb-ohci). The module i wrote to
> communicate with the device is attached in the archive and does
> basicly only translate IOCTLs into usb_control_msg() calls. My usb
> device does only communicate using the default pipe. The included test
> program (measure_update.c, sorry comments only in German) does only
> read and write 512 bytes from my device and counts the time. The speed
> was only limited by the device. With the 2.4.7 kernel I had about 60
> transfers per second. With 2.4.16 and 2.4.18 this has dropped down to
> 14.
I saw a performance problem with my Minolta Elite II scanner (never tried it
on earlier kernels - just 2.4.18 and 19). Tracked the performance problem
down to something timing out when the command length was 6. Once I padded it
to 10 the performance increased drastically. Turns out there was some timer
that would "wait" for 5 seconds before proceeding - so everything would work,
it would just take a long time. Never have found the root cause, but attached
is my kludge. Would be interested if a similar patch fixes your problem.
This is against 2.4.18 drivers/usb/hpusbscsi.c you should be able to modify
for your module. The delay showed itself taking 5 seconds in a select
statement when I did an strace (strace -t -T).
bonzo@bonzopad:/usr/src/linux$ cat otto.patch
--- a/drivers/usb/hpusbscsi.c Sat May 11 22:19:42 2002
+++ b/drivers/usb/hpusbscsi.c Sat May 11 23:19:57 2002
@@ -309,16 +309,32 @@
hpusbscsi->state = HP_STATE_BEGINNING;
TRACE_STATE;
+/*FIXME: For some reason the Minolta Elite II has a 5 second delay when the
+srb->cmd_len=6
+ * so we will pad it to 10 for now and set the length to 10 */
+
+// Begin cmd_len kludge
+ if (srb->cmd_len < MINURBCMDLEN) {
+ int i;
+
+ for (i = srb->cmd_len; i < MINURBCMDLEN; i++) // Pad the cmnd
+ {
+ srb->cmnd[i] = '\0';
+ }
+ }
+
+
/* We prepare the urb for writing out the scsi command */
FILL_BULK_URB(
&hpusbscsi->dataurb,
hpusbscsi->dev,
usb_sndbulkpipe(hpusbscsi->dev,hpusbscsi->ep_out),
srb->cmnd,
- srb->cmd_len,
+ (((int)srb->cmd_len) < MINURBCMDLEN ? MINURBCMDLEN:
+((int)srb->cmd_len)), // set length
usb_callback,
hpusbscsi
);
+// End cmd_len kludge
+
hpusbscsi->scallback = callback;
hpusbscsi->srb = srb;
--- a/drivers/usb/hpusbscsi.h Sat May 11 23:03:30 2002
+++ b/drivers/usb/hpusbscsi.h Sat May 11 23:03:45 2002
@@ -92,3 +92,6 @@
#define HP_STATE_ERROR 3 /* error has been reported */
#define HP_STATE_WAIT 4 /* waiting for status transfer */
#define HP_STATE_PREMATURE 5 /* status prematurely reported */
+
+/* FIXME: Workaround for Minolta Elite II when cmd_length=6 */
+#define MINURBCMDLEN 10
[EMAIL PROTECTED] said:
> Hi all,
> sins I upgraded from 2.4.7 to 2.4.16 my self constructed USB device
> is very slow (~ 25% of the old speed) handled by the kernel. I also
> tested this with 2.4.18 and found no difference.
> My computers are (a) a laptop with some Intel chipset (usb-uhci) and
> (b) a desktop with some VIA chipsed (usb-ohci). The module i wrote to
> communicate with the device is attached in the archive and does
> basicly only translate IOCTLs into usb_control_msg() calls. My usb
> device does only communicate using the default pipe. The included test
> program (measure_update.c, sorry comments only in German) does only
> read and write 512 bytes from my device and counts the time. The speed
> was only limited by the device. With the 2.4.7 kernel I had about 60
> transfers per second. With 2.4.16 and 2.4.18 this has dropped down to
> 14.
-----
Timothy Jedlicka, [EMAIL PROTECTED], 1-630-713-4436, AOL-IM=bonzowork
Network Entomologist, Lucent Technologies, Testers For Hire
_______________________________________________________________
Don't miss the 2002 Sprint PCS Application Developer's Conference
August 25-28 in Las Vegas -- http://devcon.sprintpcs.com/adp/index.cfm
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-users
