Chien Wong <m@xxxxxxxx> writes: >>> Shouldn't this be the same as the extra_tx_headroom set above? Not sure >>> what the +4 is for in that assignment, but it seems a bit odd to not be >>> consistent. Did you verify that an MTU of 1580 works without crashing? >>> >>> Maybe this should just be: >>> >>> hw->max_mtu = MAX_USB_WLAN_TX_PIPE_MSG_SIZE - hw->extra_tx_headroom; >>> >>> just to be sure? > > The +4 is for the header at the very beginning of the USB packet: > >/* hif_usb_send_mgmt() in hif_usb.c */ > > hdr = skb_push(skb, 4); > > *hdr++ = cpu_to_le16(skb->len - 4); > > *hdr++ = cpu_to_le16(ATH_USB_TX_STREAM_MODE_TAG); > > I suppose that the four bytes are consumed by the USB hardware and they > do not occupy buffer in the firmware. And my experiment proved this. > Unfortunately, setting MTU=1580 alone could not prevent the firmware > from crashing. The MTU only limits upper layer length, not taking MAC > overhead into account. That's why we need to take other measures such as > dropping packets before sending via USB, as proposed by my earlier > patch. Well, we can also just take the upper layer overhead into account in the MTU limit? Presumably there's a maximum MTU size that you can use without the firmware crashing? So just set the MTU limit to that :) -Toke
