On Thu, Dec 19, 2013 at 07:03:48AM +0000, Paul Zimmerman wrote: > > From: linux-usb-owner@xxxxxxxxxxxxxxx [mailto:linux-usb-owner@xxxxxxxxxxxxxxx] On Behalf Of Felipe Balbi > > Sent: Wednesday, December 18, 2013 10:03 PM > > > > On Thu, Dec 19, 2013 at 11:32:08AM +0900, Anton Tikhomirov wrote: > > > Hi, > > > > > > > Hi, > > > > > > > > On Tue, Dec 17, 2013 at 03:59:31PM +0900, Anton Tikhomirov wrote: > > > > > In accordance with specification, when sent data length is > > > > > > > > please mention section of specification. > > > > > > USB2.0 spec., 8.5.3.2 Variable-length Data Stage > > > > > > > > > > > > an exact multiple of wMaxPacketSize for the pipe and less > > > > > than requested by host, the function shall return a zero-length > > > > > packet (ZLP) to indicate the end of the Data stage to a USB host. > > > > > > > > hmm... so in USB3 mode that would be host requesting 513 bytes and us > > > > sending only 512. > > > > > > Our customer reported this issue. In their case, Windows USB2.0 host > > > requests Configuration descriptor with wLength = 255. Device replies > > > with two 64 bytes IN transactions, and stalls EP0 on 3rd IN transaction, > > > where it has to reply with ZLP. Unfortunately, we don’t have full > > > picture of what's happening on their side and why host requests > > > more bytes than actual length of Configuration descriptor. > > > > that's a bug on the host side. > > Certainly not. The host doesn't know how big the Configuration descriptor > is, so it always asks for 255 bytes (on Windows that is, I'm not sure > about Linux). yeah, that's true. Linux requests for 18 bytes (IIRC) then after that it reads wTotalLength to figure out how to request as a follow up, but the same issue could happen. > The problem here sounds like the Configuration descriptor for this > particular device is exactly 128 bytes (two packets at HS), so a 0-length > packet is needed to terminate the transfer. true -- balbi
Attachment:
signature.asc
Description: Digital signature
