Re: E1000 mbuf leaks
On 07/27/15 at 01:02P, Hans Petter Selasky wrote: > Hi, > > I'm currently doing some busdma work, and possibly stepped over some > driver bugs. When "bus_dmamap_load_mbuf_sg()" returns ENOMEM the mbuf > chain is not freed. Is there some magic in "bus_dmamap_load_mbuf_sg()" > for that error code or is there a possible memory leak in all E1000 > drivers? See attached patch. Can you open a phabricator review if this hasn't been reviewed/committed yet? cheers, Hiren pgpJXdQwQgkIL.pgp Description: PGP signature
Re: E1000 mbuf leaks
On Monday, July 27, 2015, Hans Petter Selasky wrote: > Hi, > > I'm currently doing some busdma work, and possibly stepped over some > driver bugs. When "bus_dmamap_load_mbuf_sg()" returns ENOMEM the mbuf chain > is not freed. Is there some magic in "bus_dmamap_load_mbuf_sg()" for that > error code or is there a possible memory leak in all E1000 drivers? See > attached patch. Would this explain the high mbuf usage seen on pfsense when using the igb(4) or em(4) Intel NIC drivers? https://doc.pfsense.org/index.php/Tuning_and_Troubleshooting_Network_Cards#Intel_igb.284.29_and_em.284.29_Cards Regards, Ben -- -- From: Benjamin Woods woods...@gmail.com ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
Re: E1000 mbuf leaks
On Mon, Jul 27, 2015 at 01:02:32PM +0200, Hans Petter Selasky wrote: > Hi, > > I'm currently doing some busdma work, and possibly stepped over some > driver bugs. When "bus_dmamap_load_mbuf_sg()" returns ENOMEM the mbuf > chain is not freed. Is there some magic in "bus_dmamap_load_mbuf_sg()" > for that error code or is there a possible memory leak in all E1000 > drivers? See attached patch. I don't think it's an mbuf leak since lem(4) just prepend the mbuf to the if sendq(driver will retry it later). But I think your patch looks more correct in bus_dma(9) perspective. If bus_dmamap_load_mbuf_sg(9) returned an error except EFBIG, it would be correct for lem(4) to free the mbuf chains rather than restarting the bus_dmamap_load_mbuf_sg(9) later which shall fail again with ENOMEM. ___ freebsd-current@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-current To unsubscribe, send any mail to "freebsd-current-unsubscr...@freebsd.org"
