Date: Tue, 28 Jul 2015 21:56:54 -0600 From: Scott Long <scott4long@yahoo.com> To: Dan Langille <dan@langille.org> Cc: freebsd-scsi@freebsd.org Subject: Re: dmesg output not as expected Message-ID: <27741F67-95F8-4EEF-B42F-E80626799FCB@yahoo.com> In-Reply-To: <BFD44FE6-CCF0-4973-8893-86D7E58DF7C9@langille.org> References: <5A302DC6-C3EF-4DB1-8AD1-A0314C048A28@langille.org> <BFD44FE6-CCF0-4973-8893-86D7E58DF7C9@langille.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> On Jul 28, 2015, at 8:18 PM, Dan Langille <dan@langille.org> wrote: >=20 > After rebooting this server, I see this in /var/run/dmesg. Why does = it change? >=20 > ch0 at mpt0 bus 0 scbus14 target 3 lun 0 > ch0: <COMPAQ MSL5000 Series 0520> Removable Changer SCSI-2 device > ch0: Serial Number 3G22JJP38S46 > ch0: 20.000MB/s transfers (10.000MHz DT, offset 15, 16bit) > ch0: 25 slots, 1 drive, 1 picker, 1 portal > ch0: quirks=3D0x2<NO_DVCID> >=20 > I was hoping to get more through put on this newer system with a = different card. =46rom what I see, my speed is only half of what it = was. >=20 > I'm using a PCI-E 3.0 X8 (in X16) slot. Have I missing something? >=20 The LSI controllers do Domain Validation. That means that they send a = series of test commands to each target to try to determine a safe speed = to negotiate to. The theory is that these test commands will determine = whether the target=E2=80=99s speed needs to be downshifted in order to = avoid problems with bad cables and bad connectors. It=E2=80=99s part of = the Ultra320 spec, and I think it was in the Ultra160 spec too, but was = rarely implemented there. It=E2=80=99s possible that the DV routine is = telling the card that these devices need to have their negotiated speed = reduced, and that the results it gets are a bit on-the-fence at times = and cause different observed speeds. It=E2=80=99s been a long time since I worked with DV, so my memory is a = bit fuzzy. The Adaptec hardware never did DV other than a really simple = algorithm in the BIOS for the U320 cards, and I=E2=80=99m pretty sure = that the Symbios cards pre-date DV all-together. I think that the MPT = controllers do DV in the firmware at boot, and the driver just reads the = results and uses that for the default negotiation settings. I don=E2=80=99= t see much in the MPT driver for debugging this, but I recall that the = DV algorithm that LSI used was pretty naive and overly-aggressive at = forcing down negotiation. You might want to look at what options the = LSI BIOS offers, or if you can use camcontrol to change the negotiation = up to where you think it should be. Scott
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?27741F67-95F8-4EEF-B42F-E80626799FCB>