On Fri, Feb 08, 2002, [EMAIL PROTECTED] 
<[EMAIL PROTECTED]> wrote:
> Ok, I've just completed a thorough set of tests regarding my previously
> described packet size problem with my USB ethernet (adsl) driver. It
> appears to me that either a) UHCI just can't perform like OHCI, or b)
> the UHCI driver has problems. My information is reported below. I'm
> running on RedHat 7.2 with their stock kernel - 2.4.7-10.
> 
> All transfers are on a bulk-in endpoint. URBs are queued with
> USB_QUEUE_BULK flag set. All URBs have their transfer_buffer_length set
> to 3392 bytes (the least common multiple of ATM cell size and bulk
> transfer size - 53*64). These transfers are all responses from pinging
> another machine with ping data sizes of 4k and 16k. This is the simplest
> way to illustrate and reproduce the problem. The problem affects 
> virtually all traffic, not just traffic related to pings.
> 
>   HCD     URBs  InWhole   InCATC                InURB              Ping?
> 1 UHCI    1     4717      3127 or 3074          3127 or 3074       No
> 2 UHCI    64    4717      3127 or 3074          3010 or 3063       No
> 3 OHCI    1     4717      4717                  3392 and 1325      Yes
> 4 OHCI    64    4717      4717                  3392 and 1325      Yes
> 5 OHCI    1     18868     3392*5 and variable   3392 and variable  No
> 6 OHCI    64    18868     18868                 3392*5 and 1908    Yes
> 
> Some notes on the table:
>   URBs    = number of URBs queued on the pipe
>   InWhole = entire size of data expected to receive
>   InCATC  = size of "bulk in" transfers as seen in CATC trace
>   InURB   = value of "actual_length" in the URBs returned
>   Ping?   = whether or not the ping was successful
>   
> I would expect that from a USB driver's perspective, what he gets in 
> URBs should be the same regardless of what the HCD is. Is this 
> expectation wrong?
> 
> I can accept the fact that for case #5 - we're simply running out of 
> time to receive all of the data. By the time we've queued up the URB 
> to get the last chunk of data, we've lost some data? Or, more likely, 
> we've lost some data somewhere from the first to the last chunk, but 
> since there is lots of data, the buffers are all full the urb until 
> the last one.
> 
> Anybody have any clues here? Anything else I can provide that will help?

Which UHCI driver are you using?

I assume usb-uhci since that's what RH 7.2 uses?

JE


_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to