RE: AHCI timeouts - 8.1-PRERELEASE
Quick update. Following extensive testing, I've recommended to Alexander, in a longer email, that the default for VIA's VT8251 AHCI Controller be set at AHCI_Q_NOPMP|AHCI_Q_NONCQ Use of ahci in the kernel without this option will result in timeouts, ultimately preventing the machine from performing IO. No timeouts occurred when the ahci driver was removed from the kernel. Approx 35 tests were applied, simultaneously copying a large number of files (~2000) to 8 partitions that ranged in size from 10B to 1.6GB. Testing included single and gmirror disks. The removal of NCQ resulted in a degraded performance of < 8% when compared against EDGEIS, which failed under gmirror testing. Performance under load without NCQ was better than a kernel without ahci. The test was conducted on a VIA SN18000 using two WD500GB-AAKS disks, the disks were repartitioned (gpart) and system rebooted before each test. Regards, Phil. ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "freebsd-stable-unsubscr...@freebsd.org"
Re: AHCI timeouts - 8.1-PRERELEASE
Hi. Phil wrote: > Is it expected behaviour that ahci performs a time-out during boot, even > though camcontrol indicates success? This is on 8.1-PRERELEASE csup'ed > today. > > I've enclosed dmesg, camcontrol and pciconf information. The timeout is > approximately 12 to 14 seconds for each connected disk-drive. If I > run with only usb devices, there are no timeouts. The following > dmesg is when I connect two HDD's on a VIA SN18000 motherboard, and > the BIOS has ahci selected, i.e. not IDE nor RAID. > > # pciconf -lv | grep -A 4 ahci > ah...@pci0:0:15:0: class=0x010601 card=0x62871106 chip=0x62871106 > rev=0x20 hdr=0x00 > vendor = 'VIA Technologies, Inc.' > device = 'VT8251 AHCI Controller' > class = mass storage > subclass = SATA It seems like timeout during Port Multiplier probe. I don't have such chip to test it, but I can see that Linux disables PMP and NCQ support for it. You may try it in different combinations by adding to respective line in device list in the beginning of the ahci.c quirks AHCI_Q_NOPMP and AHCI_Q_NONCQ. Instead of AHCI_Q_NONCQ (if it is needed for you) you may try to set AHCI_Q_EDGEIS. Report me please about the result. -- Alexander Motin ___ freebsd-stable@freebsd.org mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-stable To unsubscribe, send any mail to "freebsd-stable-unsubscr...@freebsd.org"