05-27-2019, 08:09 AM
(This post was last modified: 05-27-2019, 08:13 AM by bavay.
Edit Reason: Adde a few details
)
(05-24-2019, 03:57 PM)mknawabi Wrote: What did you format the filesystem as? Can you run a sequential fio write/read mix test and see if that gets it to error? What is the output of dmesg when you start seeing errors? What are the errors you described?
Cheers
Edit: Found discussion @ https://github.com/ayufan-rock64/linux-build/issues/112 that answers most of above
I've had fs either as ext4 or f2fs. I get the errors both when reading or when writing. But if i just run a simple benchmark (hdparm test), everything works. The bug is triggered when reading or writing large amount of data (a few Gb, I managed to trigger it today with ~900Mb). Depending on the kernel version, it ends up in system lockup or filesystem corruption or just that the usb drive is not reachable anymore (and can not be unmounted, so even a reboot is not going to be that much fun).
The discussion on Ayufan's github does not give any proper solution (in my mind, as I was part of this discussion, see https://github.com/ayufan-rock64/linux-b...-480827662 that contains my dmesg output when the usb crashes): the only setup that works involves two usb hubs chained to isolate the drive, this is really a very hacky workaround and not any kind of real solution (it significantly degrades the performances and a board that needs this kind of setup can not advertise being USB3!). [Edit] please keep in mind that several people tried powered usb hubs, I tried a 6A power supply for the board, I blacklisted uas (and checked that it was blacklisted), nothing helped.
As I keep on seeing more and more RK3399 boards where the USB3 is not stable under load, it seems to me that the rk3399 has a bug and is actually not able to handle usb3 properly.
PS: regarding the other issues that are answered in your link, I could not find anything... this thread was only about usb3 (and did not bring up a solution), there is still no improvement to the big.little handling in dev kernels (since it works properly with the stable kernel, there will eventually be a solution), not for the other problems. But so far, the other problems are annoying but will be fixed at some point (it might take years though) while if the rk3399 has a buggy usb3 implementation, this won't be fixable.