My experience is that if a sata controller works fine with one disk, but not 2 then there is a significant bug in the chip attempting to correctly handle 2 streams and there is likely no way to fix it because it is a hw issue. I have seen way too many controllers that must have passed a limited worthless test process with only one disk, but won't work reliably with more than one. My past testing on cheap PCI-e sata controllers was pretty ugly. They were a pretty unreliable bunch, they would generally work just good enough to make you think they work, and then fail under load. I gave up and spent a bit more to get a used HBA mode LSI SAS2008 controller but that is a pcie-8 controller and can be had for about $50. You cannot umount the filesystem once the underlying disk is gone. step #1 in umounting is to sync what data there is in ram, and with the device being gone that will never finish and hence will never umount it, and there will always be access time and/or other data to be synced from what I have seen. I have not heard of a way to tell the kernel that a device/fs is really gone and clean up the dirty writes related to that device so that it could be umounted. You can lazy umount it and that will make the mount point go away from showing up in the OS, but it really won't be umounted if there is dirty cache for the device, and/or programs with file open on the disk. On Wed, Jan 12, 2022 at 6:46 PM Aidan Walton <aidan.walton@xxxxxxxxx> wrote: > > Thanks Jani, > Gold info. I have been looking at 2 cheap Mini-PCIe adapters. One uses > the JMicron and would you believe it the other is using ASMedia. > Now I have a solid reason to choose the second rather than the first. > > In this case it will NOT be, "Better the devil you know than the > devil you don't" > and anyway I don't work hard enough to need 6Gb/s :) > ATB > Aidan > > On Thu, 13 Jan 2022 at 01:24, Jani Partanen <jiipee@xxxxxxxxxxx> wrote: > > > > Aidan Walton kirjoitti 13/01/2022 klo 2.03: > > > Hi Roger, > > > As I mentioned, it is a: > > > JMicron Technology Corp. JMB363 > > > > If my memory is correct. This chip is trash. If I am right, I have same > > chip on cheap controller what I bought to get IDE controller mainly, it > > also has 2xSATA ports and I had only troubles with that controller in linux. > > If I had no drives on SATA ports, dmesg got spammed port errors. If I > > had drives installed, they was randomly dropping in and out. Chip itself > > was running hot as hell, I mean so hot that you could not keep finger on > > chip. I did put some small finns to it and I think it did help little > > for drives dropping in and out, but didn't solve it. > > If you want a cheap controller, get this chip: > > 02:00.0 SATA controller: ASMedia Technology Inc. Device 1166 (rev 02) > > > > It's been quite stable for me and didn't cost much. Only issue I have > > that I cannot force it's ports transfer speed to example 3.0G. Spinnin > > rust do not need 6.0G speed what in my experience increase chance for > > dropouts if you cables aren't 6.0G ready and how do you tell if they are > > when cable doesn't have any mention about it? > > I think it's something to do that it's PMP card and I don't understand > > how you have to format kernel parameter when you have PMP situation. I > > can tune internal controller all ports just fine, but not this expansion > > card. Kernel tell you it's forcing them to 3.0G, but when you check > > later what transfer speed is, it's still 6.0G > > > >
