Re: EHCI_HCD resets cause USB disks to fail - caused by CPU C-states.

Re: EHCI_HCD resets cause USB disks to fail - caused by CPU C-states.

[Alan Stern]

[Adding CC's, since this is a CPU power management issue just as much 
as a USB issue.]

On Fri, 18 Sep 2009, Jan Holst Jensen wrote:

> Hi.
> 
> I recently had severe problems with a new PC. When attaching an external 
> USB disk I would see USB resets once in a while in /var/log/messages and 
> after another while the disk would disconnect and data would be lost. 
> Like this
> 
> Sep 13 13:30:56 rumleskaft kernel: [164642.760290] usb 1-5: reset high 
> speed USB device using ehci_hcd and address 3
> [repeated a number of times]
> Sep 13 13:31:17 rumleskaft kernel: [164663.430421] sd 2:0:0:0: Device 
> offlined - not ready after error recovery
> Sep 13 13:31:17 rumleskaft kernel: [164663.430459] sd 2:0:0:0: [sdb] 
> Result: hostbyte=DID_ABORT driverbyte=DRIVER_OK,SUGGEST_OK
> Sep 13 13:31:17 rumleskaft kernel: [164663.436193] usb 1-5: USB 
> disconnect, address 3
> Sep 13 13:31:17 rumleskaft kernel: [164663.436446] sd 2:0:0:0: [sdb] 
> Result: hostbyte=DID_NO_CONNECT driverbyte=DRIVER_OK,SUGGEST_OK
> Sep 13 13:31:17 rumleskaft kernel: [164663.437711] lost page write due 
> to I/O error on sdb2
> Sep 13 13:31:17 rumleskaft kernel: [164663.454184] lost page write due 
> to I/O error on sdb2
> 
> On my machine I found that disabling CPU C-states and forcing "GV3 only" 
> solved the problem completely. So in some cases it would seem that CPU 
> power management is causing timing issues (?). Can this be fixed in the 
> USB driver layer or does it take cooperation from the kernel's CPU power 
> management - or perhaps a combination of both ?
> 
> You can read more details about my issue at
> 
> http://www.fit-pc2.com/forum/viewtopic.php?f=26&t=851
> 
> Cheers
> -- Jan Holst Jensen, Denmark

For what it's worth, I haven't heard of anything like this before and I
have no idea what the answers are.  My guess is that a hardware bug is
responsible for the misbehavior while in higher C states.

Alan Stern

Re: EHCI_HCD resets cause USB disks to fail - caused by CPU C-states.

[Jan Holst Jensen]

[-- Attachment #1.1: Type: text/plain, Size: 2182 bytes --]

Alan Stern wrote:
> [Adding CC's, since this is a CPU power management issue just as much 
> as a USB issue.]
>
> On Fri, 18 Sep 2009, Jan Holst Jensen wrote:
>
>   
>> Hi.
>>
>> I recently had severe problems with a new PC. When attaching an external 
>> USB disk I would see USB resets once in a while in /var/log/messages and 
>> after another while the disk would disconnect and data would be lost. 
>> Like this
>>
>> Sep 13 13:30:56 rumleskaft kernel: [164642.760290] usb 1-5: reset high 
>> speed USB device using ehci_hcd and address 3
>> [repeated a number of times]
>> Sep 13 13:31:17 rumleskaft kernel: [164663.430421] sd 2:0:0:0: Device 
>> offlined - not ready after error recovery
>> Sep 13 13:31:17 rumleskaft kernel: [164663.430459] sd 2:0:0:0: [sdb] 
>> Result: hostbyte=DID_ABORT driverbyte=DRIVER_OK,SUGGEST_OK
>> Sep 13 13:31:17 rumleskaft kernel: [164663.436193] usb 1-5: USB 
>> disconnect, address 3
>> Sep 13 13:31:17 rumleskaft kernel: [164663.436446] sd 2:0:0:0: [sdb] 
>> Result: hostbyte=DID_NO_CONNECT driverbyte=DRIVER_OK,SUGGEST_OK
>> Sep 13 13:31:17 rumleskaft kernel: [164663.437711] lost page write due 
>> to I/O error on sdb2
>> Sep 13 13:31:17 rumleskaft kernel: [164663.454184] lost page write due 
>> to I/O error on sdb2
>>
>> On my machine I found that disabling CPU C-states and forcing "GV3 only" 
>> solved the problem completely. So in some cases it would seem that CPU 
>> power management is causing timing issues (?). Can this be fixed in the 
>> USB driver layer or does it take cooperation from the kernel's CPU power 
>> management - or perhaps a combination of both ?
>>
>> You can read more details about my issue at
>>
>> http://www.fit-pc2.com/forum/viewtopic.php?f=26&t=851
>>
>> Cheers
>> -- Jan Holst Jensen, Denmark
>>     
>
> For what it's worth, I haven't heard of anything like this before and I
> have no idea what the answers are.  My guess is that a hardware bug is
> responsible for the misbehavior while in higher C states.
>
> Alan Stern
>   

Could be. Hardware info for my system can be found at 
http://www.jan-holst.dk/linux-usb/sysconfig.txt. It is the output of 
/proc/version, /etc/issue and lshw.

-- Jan

[-- Attachment #1.2: Type: text/html, Size: 2683 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: EHCI_HCD resets cause USB disks to fail - caused by CPU C-states.

[Jan Holst Jensen]

Sorry - posted first as text + HTML. Re-posting as plain text.

Alan Stern wrote:
> [Adding CC's, since this is a CPU power management issue just as much 
> as a USB issue.]
>
> On Fri, 18 Sep 2009, Jan Holst Jensen wrote:
>
>   
>> Hi.
>>
>> I recently had severe problems with a new PC. When attaching an external 
>> USB disk I would see USB resets once in a while in /var/log/messages and 
>> after another while the disk would disconnect and data would be lost. 
>> Like this
>>
>> Sep 13 13:30:56 rumleskaft kernel: [164642.760290] usb 1-5: reset high 
>> speed USB device using ehci_hcd and address 3
>> [repeated a number of times]
>> Sep 13 13:31:17 rumleskaft kernel: [164663.430421] sd 2:0:0:0: Device 
>> offlined - not ready after error recovery
>> Sep 13 13:31:17 rumleskaft kernel: [164663.430459] sd 2:0:0:0: [sdb] 
>> Result: hostbyte=DID_ABORT driverbyte=DRIVER_OK,SUGGEST_OK
>> Sep 13 13:31:17 rumleskaft kernel: [164663.436193] usb 1-5: USB 
>> disconnect, address 3
>> Sep 13 13:31:17 rumleskaft kernel: [164663.436446] sd 2:0:0:0: [sdb] 
>> Result: hostbyte=DID_NO_CONNECT driverbyte=DRIVER_OK,SUGGEST_OK
>> Sep 13 13:31:17 rumleskaft kernel: [164663.437711] lost page write due 
>> to I/O error on sdb2
>> Sep 13 13:31:17 rumleskaft kernel: [164663.454184] lost page write due 
>> to I/O error on sdb2
>>
>> On my machine I found that disabling CPU C-states and forcing "GV3 only" 
>> solved the problem completely. So in some cases it would seem that CPU 
>> power management is causing timing issues (?). Can this be fixed in the 
>> USB driver layer or does it take cooperation from the kernel's CPU power 
>> management - or perhaps a combination of both ?
>>
>> You can read more details about my issue at
>>
>> http://www.fit-pc2.com/forum/viewtopic.php?f=26&t=851
>>
>> Cheers
>> -- Jan Holst Jensen, Denmark
>>     
>
> For what it's worth, I haven't heard of anything like this before and I
> have no idea what the answers are.  My guess is that a hardware bug is
> responsible for the misbehavior while in higher C states.
>
> Alan Stern
>   

Could be. Hardware info for my system can be found at 
http://www.jan-holst.dk/linux-usb/sysconfig.txt. It is the output of 
/proc/version, /etc/issue and lshw.

-- Jan