Bad news: Jansen's solution seems to delay the problem, but not fix it for me.
I got a new 500GB drive which dies when I try this:
# mke2fs -c /dev/sda1
The block at which it dies depends on whether I run a kernel with Jansen's suggestion or not.
With USB_EHCI_SPLIT_ISO, USB_EHCI_ROOT_HUB_IT or USB_EHCI_TT_NEWSCHED enabled and USB_BANDWITH disabled, it dies pretty soon.
On the other hand, when USB_EHCI_SPLIT_ISO, USB_EHCI_ROOT_HUB_IT and USB_EHCI_TT_NEWSCHED are disabled and USB_BANDWITH is enabled, it takes
as much as 10 times longer for the USB connection to die, but it dies at the end all the same.
It is definitely a bug in the handling of USB controllers. I gathered some evidence that this is the case using a different machine and a 300 GB disk. I have two different generic
USB enclosures. The same disk mounted on one of them works perfectly, while mounted on the other it exhibits the connection problems and eventually dies.
Finally, a caveat: Creating a partition in the 500GB disk without badblock checking worked fine. I could also copy some large movie files into the disk, but when it eventually died,
it killed the partition. The data in the disk was totally destroyed and I could not recover it. Maybe this is a particularly bad controller (reported as "Samsung" by sysfs) but the
severing of the USB bus connection should never have the effect of totally destroying all the data stored in a device. Too bad.
Bad news: Jansen's solution seems to delay the problem, but not fix it for me.
I got a new 500GB drive which dies when I try this:
# mke2fs -c /dev/sda1
The block at which it dies depends on whether I run a kernel with Jansen's suggestion or not. ROOT_HUB_ IT or USB_EHCI_ TT_NEWSCHED enabled and USB_BANDWITH disabled, it dies pretty soon. ROOT_HUB_ IT and USB_EHCI_ TT_NEWSCHED are disabled and USB_BANDWITH is enabled, it takes
With USB_EHCI_SPLIT_ISO, USB_EHCI_
On the other hand, when USB_EHCI_SPLIT_ISO, USB_EHCI_
as much as 10 times longer for the USB connection to die, but it dies at the end all the same.
It is definitely a bug in the handling of USB controllers. I gathered some evidence that this is the case using a different machine and a 300 GB disk. I have two different generic
USB enclosures. The same disk mounted on one of them works perfectly, while mounted on the other it exhibits the connection problems and eventually dies.
Finally, a caveat: Creating a partition in the 500GB disk without badblock checking worked fine. I could also copy some large movie files into the disk, but when it eventually died,
it killed the partition. The data in the disk was totally destroyed and I could not recover it. Maybe this is a particularly bad controller (reported as "Samsung" by sysfs) but the
severing of the USB bus connection should never have the effect of totally destroying all the data stored in a device. Too bad.