On 2021/03/03 15:10, Karel Gardas wrote: > > On 3/2/21 11:09 PM, Mark Schneider wrote: > > > > It looks like "not so perfect" mix of a bit outdated ASUS mainboard / > > it's BIOS, less old Samsung PRO 860 512GB or 1TB SSD drives and the > > handlig of I/O errors in OpenBSD 6.8 or 6.9beta. > > > > bios0 at mainbus0: SMBIOS rev. 2.7 @ 0xdc312018 (61 entries) > > bios0: vendor American Megatrends Inc. version "2201" date 03/23/2015 > > bios0: ASUSTeK COMPUTER INC. CROSSHAIR V FORMULA-Z > > acpi0 at bios0: ACPI 5.0 > > > > The Samsung PRO 860 SSDs are new so I do not expect the problem there > > (as they are working on Linux). > > > > I have taken two of those Samsung PRO 860 512GB SSD drives and connected > > them to another "P8B WS, BIOS 0704 07/25/2011" Xeon based Asus mainboard > > and there are no AHCI errors showing up. I have tested there isolated > > drives as well as plain RAID1 and ecnrypted RAID1 (nested, not using the > > "-c 1C" option of bioctl in OpenBSD 6.9beta) writing 1, 2 10 or 20GBytes > > big files to the RAID device and there was no issue at all. > > > > The issue is showing up on three Asus FX CPU based mainboards ( 1 x > > "SABERTOOTH 990FX R2.0" and 2 x "CROSSHAIR V FORMULA-Z" - all of them > > have the same SB950 chipset) > > > > AMD ® SB950 Chipset > > Supports AMD ® QUAD-GPU CrossFireXTM Technology > > - 6 x SATA 6.0 Gb/s ports with RAID 0, 1, 5 and 10 support > > > > Looks like you are not the only one having issue with AMD chipset and new > samsung SSD: > > https://eu.community.samsung.com/t5/computers-it/860-evo-250gb-causing-freezes-on-amd-system/td-p/575813/page/2 > https://community.amd.com/t5/drivers-software/990fx-sb950-chipset-drivers-not-working-with-samsung-860-pro-ssd/td-p/100175 > > And here you even have a trace of issues in Linux kernel bugzilla: > > https://bugzilla.kernel.org/show_bug.cgi?id=201693 > > If you read all info on those pages, you will find some workarounds for you > to test, but generally speaking I'm afraid this hardware combination > does not seem to be too trustful... >
"So I guess we should consider doing a kernel side quirk where the kernel disables NCQ on the combination of having a Samsung 860 or 870 SSD with a SATA controller on these older AMD chipsets. This does require having a list of PCI-ids for the controllers on which to enable this quirk." ouch