Date: Thu, 5 Dec 2013 22:05:16 +0100 From: Michael Tuexen <Michael.Tuexen@lurchi.franken.de> To: Adrian Chadd <adrian@freebsd.org> Cc: Yong-Hyeon Pyun <pyunyh@gmail.com>, Jack F Vogel <jfv@freebsd.org>, "freebsd-net@freebsd.org list" <freebsd-net@freebsd.org> Subject: Re: A small fix for if_em.c, if_igb.c, if_ixgbe.c Message-ID: <B89B1E2D-BAF0-4815-B3AB-EB226F4F76DE@lurchi.franken.de> In-Reply-To: <CAJ-Vmo=kfoPMYjZ0WAtqmoJMz1utXH50SW9N92RA83EMUzY7WA@mail.gmail.com> References: <521B9C2A-EECC-4412-9F68-2235320EF324@lurchi.franken.de> <20131202022338.GA3500@michelle.cdnetworks.com> <B9593E83-E687-49E9-ABDC-B2DD615180E9@lurchi.franken.de> <20131203021658.GC2981@michelle.cdnetworks.com> <CAJ-Vmo=kfoPMYjZ0WAtqmoJMz1utXH50SW9N92RA83EMUzY7WA@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Dec 5, 2013, at 7:29 PM, Adrian Chadd <adrian@freebsd.org> wrote: > Hi, > > Yes. Looking at the ixgbe code, ixgbe_mq_start_locked() returns an > error from ixgbe_xmit() but if it fails, it puts the buffer back. But > it's already successfully queued a frame to the driver, so in this > instance it shouldn't return the error from ixgbe_mq_start_locked(). > > The same deal in if_em.c and igb.c > > Now, drbr_putback() used to fail and now it doesn't, as you've said. > So we should change the xxx_mq_start_locked() to set err=0 if we go > via the drbr_putback() routine, as it hasn't actually failed to > transmit. > > Now the very dirty thing is this - the error from xxx_transmit() is > for the mbuf being queued at the end; but xxx_mq_start_locked() > failures are for transmitting from the front. If there's only packet > in the queue and that fails then they're the same thing and returning > the error from xxx_mq_start_locked() matches the current mbuf being > queued. But otherwise, they're referring to totally different packets. > For TCP this may hurt; the TCP stack treats ENOBUFS a certain way and > kicks off a timer to schedule a retransmit. I don't think we can fix > _this_ right now. Just to be clear: This would mean that xxx_transmit() would return an error even if the packet provided in the call xxx_transmit() is enqueued and not dropped? This would also be problem with the current SCTP stack. Best regards Michael > > So Michael - can you redo your patch to set err=0 if drbr_putback() is > called, and retest? > > Thanks! > > > > > -adrian >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?B89B1E2D-BAF0-4815-B3AB-EB226F4F76DE>