Date: Thu, 8 Mar 2012 11:38:09 -0800 From: Adrian Chadd <adrian@freebsd.org> To: John Baldwin <jhb@freebsd.org> Cc: freebsd-current@freebsd.org, freebsd-wireless@freebsd.org, Bernhard Schmidt <bschmidt@freebsd.org> Subject: Re: patches for if_iwi and wlan for WEP mode Message-ID: <CAJ-Vmok7gd9f5V1DaYfUPU4ic98TzPTvnJCjOqfV-y21V6xDyQ@mail.gmail.com> In-Reply-To: <201203081043.44801.jhb@freebsd.org> References: <20120306.024212.108736612.iwasaki@jp.FreeBSD.org> <201203072017.44409.bschmidt@freebsd.org> <CAJ-Vmo=T4Muzsgafhfq2QLencWdr37-frpQeczn_U4wx5MhwxA@mail.gmail.com> <201203081043.44801.jhb@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On 8 March 2012 07:43, John Baldwin <jhb@freebsd.org> wrote: > However, you could do that by having a net80211_ifattach() type thing tha= t > sets if_transmit and invokes the driver-provided if_start. =A0I don't thi= nk > wireless devices are using multiple transmit queues in such a way that > if_transmit would be a benefit, and if_start is a simpler model. THe problem is that the default if_transmit uses IFQ_ENQUEUE and the _start() versions out there use IFQ_DEQUEUE without necessarily handling fragments at all. We end up having m->m_nextpkt NULL'ed upon IFQ_ENQUEUE, thus not only making a fragment list entirely broken, the fragments themselves are actually leaked. So we end up running out of mbufs. The solution I can see working at the moment is creating if_transmit for each wireless device and correctly handling any fragments in the head mbuf before calling IFQ_DEQUEUE. Adrian
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAJ-Vmok7gd9f5V1DaYfUPU4ic98TzPTvnJCjOqfV-y21V6xDyQ>