Mike Andrews
2005-Dec-19 18:10 UTC
"Request Requeued/Retrying Command" with twa card on FreeBSD 6.0
I've got a weird but apparently minor issue with a 3Ware 9500S-4LP in a FreeBSD 6.0-RELEASE system... When the system is run with boot.verbose="YES" in /boot/loader.conf, I get these messages on the console: (da0:twa0:0:0:0): Request Requeued (da0:twa0:0:0:0): Retrying Command (da0:twa0:0:0:0): Request Requeued (da0:twa0:0:0:0): Retrying Command (repeat a few dozen times each time) ...during some heavy disk activity, like, for example, qpopper rewriting a 50 MB mailbox. (I know qpopper's a bit of a slug.) Not all heavy disk IO triggers it, though; I can do something like "dd if=/dev/zero of=/tmp/testfile bs=10485760 count=128" followed by "dd if=/tmp/testfile of=/dev/null bs=10485760" and have it complete without a single SCSI timeout message. Turn boot.verbose="NO" off and the messages go away, but IO still lags badly. "camcontrol tags" says it has 254 tag openings, which seems like a bit much. Dropping down to 16 seems to make the messages go away. Do the twa cards even do TCQ between the card and the OS? From what I understand, they won't do NCQ between the card and the SATA disks, even if you have something like Barracudas that support it -- you need the new 3Ware 9550 card for that. If they don't, why does it advertise 254 tags, and if it does, why does it seem to choke on the default setting of 254? Other than laggy IO (shell prompts hang, MP3's skip for 5 - 30 seconds) there aren't any actual IO errors or anything else wrong, which for this particular system is just minor annoyance instead of serious problem. 5.4-RELEASE did the same thing. I have an apparent workaround, I'm just trying to sanity-check it and its necessity. This is an Asus P4C800-E Deluxe board, so I'm running the twa card in a 32-bit slot. The system is otherwise totally idle when this happens, no interrupt storms, USB and Firewire disabled in the BIOS so we don't have IRQs shared with Giant-locked drivers... dmesg, vmstat -i, other stuff available on request if needed. Kernel config is the stock 'SMP' one. Mike Andrews * mandrews@bit0.com * http://www.bit0.com It's not news, it's Fark.com. Carpe cavy!
Vinod Kashyap
2005-Dec-19 18:41 UTC
"Request Requeued/Retrying Command" with twa card on FreeBSD 6.0
> -----Original Message----- > From: owner-freebsd-stable@freebsd.org > [mailto:owner-freebsd-stable@freebsd.org] On Behalf Of Mike Andrews > Sent: Monday, December 19, 2005 6:10 PM > To: freebsd-stable@freebsd.org > Subject: "Request Requeued/Retrying Command" with twa card on > FreeBSD 6.0 > > I've got a weird but apparently minor issue with a 3Ware > 9500S-4LP in a FreeBSD 6.0-RELEASE system... > > When the system is run with boot.verbose="YES" in > /boot/loader.conf, I get these messages on the console: > > (da0:twa0:0:0:0): Request Requeued > (da0:twa0:0:0:0): Retrying Command > (da0:twa0:0:0:0): Request Requeued > (da0:twa0:0:0:0): Retrying Command > (repeat a few dozen times each time) > > ...during some heavy disk activity, like, for example, > qpopper rewriting a 50 MB mailbox. (I know qpopper's a bit > of a slug.) Not all heavy disk IO triggers it, though; I can > do something like "dd if=/dev/zero of=/tmp/testfile > bs=10485760 count=128" followed by "dd if=/tmp/testfile > of=/dev/null bs=10485760" and have it complete without a > single SCSI timeout message. > > Turn boot.verbose="NO" off and the messages go away, but IO > still lags badly. > > "camcontrol tags" says it has 254 tag openings, which seems > like a bit much. Dropping down to 16 seems to make the > messages go away. > > Do the twa cards even do TCQ between the card and the OS? > From what I understand, they won't do NCQ between the card > and the SATA disks, even if you have something like > Barracudas that support it -- you need the new 3Ware 9550 > card for that. > > If they don't, why does it advertise 254 tags, and if it > does, why does it seem to choke on the default setting of 254? >Although the driver can handle 254 simultaneous requests, if the I/O's, are big, there's the possibility that the firmware cannot accept new requests even if 254 has not been hit. In that case, the driver will have to ask CAM to requeue the requests.> Other than laggy IO (shell prompts hang, MP3's skip for 5 - > 30 seconds) there aren't any actual IO errors or anything > else wrong, which for this particular system is just minor > annoyance instead of serious problem. > 5.4-RELEASE did the same thing. I have an apparent > workaround, I'm just trying to sanity-check it and its necessity. > > This is an Asus P4C800-E Deluxe board, so I'm running the twa > card in a 32-bit slot. The system is otherwise totally idle > when this happens, no interrupt storms, USB and Firewire > disabled in the BIOS so we don't have IRQs shared with > Giant-locked drivers... > > dmesg, vmstat -i, other stuff available on request if needed. > Kernel config is the stock 'SMP' one. > > > > Mike Andrews * mandrews@bit0.com * http://www.bit0.com > It's not news, it's Fark.com. Carpe cavy! > > _______________________________________________ > freebsd-stable@freebsd.org mailing list > http://lists.freebsd.org/mailman/listinfo/freebsd-stable > To unsubscribe, send any mail to > "freebsd-stable-unsubscribe@freebsd.org" >-------------------------------------------------------- CONFIDENTIALITY NOTICE: This e-mail message, including any attachments, is for the sole use of the intended recipient(s) and contains information that is confidential and proprietary to Applied Micro Circuits Corporation or its subsidiaries. It is to be used solely for the purpose of furthering the parties' business relationship. All unauthorized review, use, disclosure or distribution is prohibited. If you are not the intended recipient, please contact the sender by reply e-mail and destroy all copies of the original message.