[haiku-bugs] Re: [Haiku] #9311: usb_midi fix to handle more "Class-Compliant" devices
- From: "mmlr" <trac@xxxxxxxxxxxx>
- Date: Wed, 09 Jan 2013 22:55:31 -0000
#9311: usb_midi fix to handle more "Class-Compliant" devices
---------------------------+-------------------------
Reporter: Pete | Owner: mmlr
Type: bug | Status: new
Priority: normal | Milestone: R1
Component: Drivers/USB | Version: R1/alpha4.1
Resolution: | Keywords:
Blocked By: | Blocking:
Has a Patch: 1 | Platform: All
---------------------------+-------------------------
Comment (by mmlr):
Replying to [comment:4 phoudoin]:
> I wonder if in fact the root issue is not deeper, in the USB stack
itself. There is absolutely no reason to pass the bulk request with 2048
size while the stack knows that this endpoint only support 4 bytes max per
request.
That is not how it works. The max packet size is just that, the maximum
size of each packet. A request will be fulfilled using as many packets as
is required to move the amount of data it is given. The stack splits these
buffers correctly into multiple packets internally. It's an implementation
detail if you will. It allows a device to work with a fixed and small
amount of storage if needed.
> Clamping it in usb_midi is one way, but maybe it's in the stack itself
that it should be done. In particular because the USB stack's queue_bulk()
never required to pass a buffer no bigger than the endpoint's max packet
size...
That is right, the queue_bulk() call does not have buffer size
restrictions. The usb_disk driver does transfer large amounts of data
using single calls to queue_bulk() for example, something that would
otherwise be really impractical.
The device classes usually have specifications that tell what kind of
limit requests as a whole should have, or in what manner these should be
communicated. Some have generic descriptors, some may use the max packet
size, some use requests, others use higher level protocols to determine
these limits.
Looking at the specs the packet sizes aren't specified. Looking at the
linux usb midi driver (http://lxr.linux.no/linux+v3.7.1/sound/usb/midi.c)
they use the max packet size as the max transfer size by default. It is
also clear from looking at this driver that there are quite a few broken
devices and a lot of vendor specific protocols that are slightly
different.
So using the max packet size probably makes sense. Using such short
requests does impact performance of course as the overhead per request is
quite considerable. However devices that allow for larger buffers will
probably advertise them and lessen this issue. It is not a big issue
overall, as for example USB HID transfers are usually in the "couple of
bytes" range as well (though these are usually not streaming either).
--
Ticket URL: <http://dev.haiku-os.org/ticket/9311#comment:5>
Haiku <http://dev.haiku-os.org>
Haiku - the operating system.
Other related posts:
