Thanks Alan. Responses inline below. > If the hub is plugged directly into an EHCI controller and there's no > companion USB-1.1 controller (as is the case with Intel's > "rate-matching" hubs), there's no way to do it at all. Unfortunatley yes that seems to be exactly the case. Disabling ehci-hcd was the first thing I tried, but then the USB devices are not enumerated by ohci/uhci. >> ehci_hcd 0000:00:1a.0: HC died; cleaning up > Oddly enough, this error message has nothing to do with usb-storage. > It refers to the periodic schedule, which is used for things like hubs, input devices, audio & video, etc. Thanks for the clarification. It definitely seems to coincide with when the USB storage disconnects. But I guess that makes sense : the hub is internal to the server, so the kb/mouse hang off it also. If that hub (or the ehci-hcd itself) is reset then the usb-storage would also be disconnected. > Include more of the context After the USB storage disconnect, I do see the usb-storage device reconnect almost immediately. But filesystems mounted at the time of the disconnect become inaccessible; the USB storage device comes back as a new sd entry. e.g. if it was sda before the reset then it will come back as sdb, and the filesystems on sda are then no longer accessible. Eventually ext3 gives up on the dead sda and panics. > Can you build a kernel with CONFIG_USB_DEBUG enabled and post the dmesg log from a failure? I will try. The issue is not easy to reproduce, but it does occur on multiple servers so I'll try to collate more data. Unfortunately, we use the HP as an embedded server and that USB storage is where the root filesystem and logs are written to, so when it disconnects we also lose the kernel logs etc. But I'll set up a serial console to try and capture the kernel kernel debug messages remotely. For reference, we have about 30 of the HP DL120 G7 servers, of which at least 4 have shown the USB disconnect problem. The same USB storage modules are rock-solid on the previous generation HP DL120 G6 servers. Thought: I wonder if there is a way to leverage the USB suspend functionality, so that the device reconnects back in as the existing sda, instead of as a new device... that might be sufficient to mask the filesystem problem even if it doesn't address why the hub is being reset. -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
