blktests nvme/039 failure with kernel v6.1-rc5

blktests nvme/039 failure with kernel v6.1-rc5

[Shinichiro Kawasaki]

I found that the test case nvme/039 fails with the kernel v6.1-rc5. The test
case tests "nvme error logging by injecting errors". The failure message was as
follows:


nvme/039 => nvme5n1 (test error logging)                     [failed]
    runtime    ...  0.402s
    --- tests/nvme/039.out      2022-11-14 16:40:48.598495709 +0900
    +++ /home/shin/kts/kernel-test-suite/sets/blktests/log/runlog/nvme5n1/nvme/039.out.bad      2022-11-14 17:14:05.387628216 +0900
    @@ -2,6 +2,4 @@
      Read(0x2) @ LBA 0, 1 blocks, Unrecovered Read Error (sct 0x2 / sc 0x81) DNR
      Read(0x2) @ LBA 0, 1 blocks, Unknown (sct 0x3 / sc 0x75) DNR
      Write(0x1) @ LBA 0, 1 blocks, Write Fault (sct 0x2 / sc 0x80) DNR
    - Identify(0x6), Access Denied (sct 0x2 / sc 0x86) DNR
    - Unknown(0x96), Invalid Command Opcode (sct 0x0 / sc 0x1) DNR
     Test complete


I checked changes in rc5 and found that the commit d7ac8dca938c ("nvme: quiet
user passthrough command errors") triggers the failure. By reverting the commit,
the failure disappeared. I guess the test case needs modification to match with
the trigger commit.

Alan, Keith, could you take a look in this?

-- 
Shin'ichiro Kawasaki

Re: blktests nvme/039 failure with kernel v6.1-rc5

[Alan Adamson]

> On Nov 14, 2022, at 6:29 PM, Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com> wrote:
> 
> I found that the test case nvme/039 fails with the kernel v6.1-rc5. The test
> case tests "nvme error logging by injecting errors". The failure message was as
> follows:
> 
> 
> nvme/039 => nvme5n1 (test error logging)                     [failed]
>    runtime    ...  0.402s
>    --- tests/nvme/039.out      2022-11-14 16:40:48.598495709 +0900
>    +++ /home/shin/kts/kernel-test-suite/sets/blktests/log/runlog/nvme5n1/nvme/039.out.bad      2022-11-14 17:14:05.387628216 +0900
>    @@ -2,6 +2,4 @@
>      Read(0x2) @ LBA 0, 1 blocks, Unrecovered Read Error (sct 0x2 / sc 0x81) DNR
>      Read(0x2) @ LBA 0, 1 blocks, Unknown (sct 0x3 / sc 0x75) DNR
>      Write(0x1) @ LBA 0, 1 blocks, Write Fault (sct 0x2 / sc 0x80) DNR
>    - Identify(0x6), Access Denied (sct 0x2 / sc 0x86) DNR
>    - Unknown(0x96), Invalid Command Opcode (sct 0x0 / sc 0x1) DNR
>     Test complete
> 
> 
> I checked changes in rc5 and found that the commit d7ac8dca938c ("nvme: quiet
> user passthrough command errors") triggers the failure. By reverting the commit,
> the failure disappeared. I guess the test case needs modification to match with
> the trigger commit.
> 
> Alan, Keith, could you take a look in this?


The patch that quiets logging for passthrough commands caused this.  We just need to remove the last 3 output lines from 039.out and that will resolve it until we add a logging opt-in feature for userland/passthrough commands.

Can can submit a patch to change 039.out.

I’m also working on a opt-in feature.

Alan

Re: blktests nvme/039 failure with kernel v6.1-rc5

[Chaitanya Kulkarni]

I checked changes in rc5 and found that the commit d7ac8dca938c ("nvme: 
quiet
>> user passthrough command errors") triggers the failure. By reverting the commit,
>> the failure disappeared. I guess the test case needs modification to match with
>> the trigger commit.
>>
>> Alan, Keith, could you take a look in this?
> 
> 
> The patch that quiets logging for passthrough commands caused this.  We just need to remove the last 3 output lines from 039.out and that will resolve it until we add a logging opt-in feature for userland/passthrough commands.
> 
> Can can submit a patch to change 039.out.
> 

It will be great if we can fix this as early as we can before we start 
getting flood of similar reports on lkml.

-ck

Re: blktests nvme/039 failure with kernel v6.1-rc5

[Alan Adamson]

> On Nov 15, 2022, at 11:36 AM, Chaitanya Kulkarni <chaitanyak@nvidia.com> wrote:
> 
> I checked changes in rc5 and found that the commit d7ac8dca938c ("nvme: 
> quiet
>>> user passthrough command errors") triggers the failure. By reverting the commit,
>>> the failure disappeared. I guess the test case needs modification to match with
>>> the trigger commit.
>>> 
>>> Alan, Keith, could you take a look in this?
>> 
>> 
>> The patch that quiets logging for passthrough commands caused this.  We just need to remove the last 3 output lines from 039.out and that will resolve it until we add a logging opt-in feature for userland/passthrough commands.
>> 
>> Can can submit a patch to change 039.out.
>> 
> 
> It will be great if we can fix this as early as we can before we start 
> getting flood of similar reports on lkml.

I’ll work on a patch.

Alan