Re: [PATCH] cleanup usb_platform_finddevs()

[Guy Harris <guy () alum mit edu>]

Abeni Paolo wrote:

> the attached patch cleanup the usb_platform_finddevs(), removing the
> dependency from debugfs. There are also some other minor cleanup in the
> pcap-usb-linux.c file (unused variables removed and indentation fix).

Checked in.

> I plan to post the aligment fix for usb header in a separate patch.
> The code is ready, but I feel uneasy about it. If I set the header aling
> in the sf_next_packet(), this will be the first data link to use such an
> hack, right?!?

Yes, but it's not the first data link that would need it.  The only 
reason we don't do it for the Prism header is that we currently don't 
dissect the Prism header.

> Personally I will prefer to enforce the align (little
> endian) at capture time, but what do you think? 
>
> Just to give some figures, I think that the performance gain related
> to memory mapped access (which should be disabled if the byte ordering
> is enforced at capture time) should be roughly 20%-30%. It should be
> greater when the data block exchanged on the usb bus are relativly
> small, but in such condition the bus bandwith utilization is very
> sub-optimal.

I'd say a 20-30% performance gain is worth a hack in the 
capture-file-reading code path.
-
This is the tcpdump-workers list.
Visit https://cod.sandelman.ca/ to unsubscribe.