On Wed, 13 Feb 2002, Greg KH wrote: > > Yes, I'm referring to 7-29. I don't have the source handy and my > > "information" is anecdotal. It might be a problem with a particular > > implementation that is not the "standard" USB package for Linux. > > > > If you say it's there, I'll believe you. On the other hand, if some kind > > person would send me the source for hub.c I could check for myself. ;-) > > It's available for all at http://www.kernel.org/ :)
Hi Greg and David, During my ongoing investigation wrt. to the usb_reset_device issue I just happened to look into this code during the last hour or so. Yes, I do totally agree with David: there is absolutely _no_ debounce interval (connect-detect-to-port-reset) implemented in 2.4.18-pre9 hub.c According to USB 7.1.7.1 this is mandatory and must be TATTDB > 100 ms. It is also shown in fig. 7-19 of the USB 1.1 specs (I'm validating vs. 1.1). The mentioned 400ms delay for low speed devices is just were the debounce delay should be. I've just done a simple try and yes, this solves an "device not accepting address..." issue for me: when I connect a 400 mA buspowered device to the cold port (in contrast to have it already conneted before insmod usbcore which enables the ports independent from connect events) I always got the above message. In most cases one of the retries succeded later on anyway. Admittedly, this might be some device problem or root hub power instability - but implementing the debounce delay solves this (150-200 ms were required in this case). Btw. another concern wrt. to this code from hub.c is, there are several places calling usb_get_port_status passing a pointer to a stack-located struct usb_port_status - which becomes the transfer_buffer for the usb_control_msg) call. Mapping stack locations for PCI-DMA however is invalid. I'm wondering whether we could simply have shared buffer for this allocated from khubd instead of spending much effort to alloc/check/free such stuff at many places. Martin _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
