This made me recall having to download and alter a Windows SATA Host Controller driver. For some reason I had to get hold of an 6Gbyte development .iso to do this (that's Microsoft for you)... Now that was fun and games!! Basically because the SATA controller driver tried to enable NCQ (I can't remember which now!!) for some old Hitachi IDE drives - I had hooked these up via a SATA-PATA bridge - "the bridge said NCQ - YES!! - the drive said NCQ - NO!!". The drives would lock up Windows when I hooked them up to the controller and start "twanging" constantly as they were reset, over and over again...
I read through that bug report - interesting. It shouldn't be too hard for you add a User patch for your Gentoo kernel - to force-ably disable the attempt to negotiate SATA 3.1 and T13 ATA ACS-3 support - for the SATA connection. Since (I guess) you are building the kernel from source anyway! Basically your 850 EVO returns "RECEIVE/SEND FPDMA QUEUED supported" when initially queried Then when the (Linux) kernel tries to actually queue these commands - the 850 EVO firmware says "Uhhhmm, duh - no I don't know how to do that one"!! Just out of interest what make is the Host Controller on your motherboard... Is it a Intel one? Or some crappy addon chipset? Perhaps you could post the output of lscpi (with lots of -vvvv flags - just the Host Controller bit)? I'm starting get the feeling that Samsung take some very caliver shortcuts with the QC on their SSDs. I've recently bought a Samsung 850 Pro - sucker - as I have already experienced poor after-sales support with some Samsung 830 SSD "issues". Now I've found out that the 850 Pro also doesn't support non-deterministic trim - so I can't use it with *any* Host Controllers with LSI-based firmware (or whatever they are called now) - because trim support will be completely disabled!! Sure we all want V-NAND - but perhaps I should have waited a year or two for Intel to catch up... To quote from the bug report (referring to the crappy Samsung 840 EVO firmware): >> I tried to contact Samsung support, but they answered that "for the support of this product, you should contact the reseller". >> In other words, they don't want to answer. >>We are dealing with a drive that claims (during identification) to have a capability that it doesn't actually have. :-( On 25 February 2015 at 12:45, Stefan G. Weichinger <li...@xunil.at> wrote: > On 25.02.2015 10:23, Stefan G. Weichinger wrote: > > > good hint, thanks! I will report back if I find something. > > Swapped the cable and also the SATA-socket on the board. > > It always gives the same result ... > > What is interesting: the HDDs negotiate their NCQ fine: > > > [ 2.254804] ata2.00: ATA-8: Hitachi HDS721010CLA632, JP4OA41A, max > UDMA/133 > [ 2.254809] ata2.00: 1953525168 sectors, multi 16: LBA48 NCQ (depth > 31/32), AA > > [ 2.254842] ata5.00: ATA-8: ST31000524AS, JC4B, max UDMA/133 > [ 2.254846] ata5.00: 1953525168 sectors, multi 16: LBA48 NCQ (depth > 31/32) > > and that with those cheap cables ... > > I googled that and found > > https://bugzilla.kernel.org/show_bug.cgi?id=72341 > > digging through that now :-) > > > -- All the best, Robert
