On 11/3/21 3:47 PM, Keith Busch wrote:
> On Wed, Nov 03, 2021 at 02:38:53PM -0700, Keith Busch wrote:
>> On Wed, Nov 03, 2021 at 01:51:18PM -0600, Jens Axboe wrote:
>>> On 11/3/21 8:14 AM, kernel test robot wrote:
>>>>
>>>>
>>>> Greeting,
>>>>
>>>> FYI, we noticed the following commit (built with gcc-9):
>>>>
>>>> commit: f9c499bbbf603389abad60d1931c16b2f96dee06 ("[PATCH 1/2] nvme: move command clear into the various setup helpers")
>>>> url: https://github.com/0day-ci/linux/commits/Jens-Axboe/nvme-move-command-clear-into-the-various-setup-helpers/20211018-214956
>>>> base: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git 519d81956ee277b4419c723adfb154603c2565ba
>>>> patch link: https://lore.kernel.org/linux-block/20211018124934.235658-2-axboe@xxxxxxxxx
>>>>
>>>> in testcase: will-it-scale
>>>> version: will-it-scale-x86_64-a34a85c-1_20211029
>>>> with following parameters:
>>>>
>>>> nr_task: 50%
>>>> mode: process
>>>> test: readseek1
>>>> cpufreq_governor: performance
>>>> ucode: 0x700001e
>>>>
>>>> test-description: Will It Scale takes a testcase and runs it from 1 through to n parallel copies to see if the testcase will scale. It builds both a process and threads based test in order to see any differences between the two.
>>>> test-url: https://github.com/antonblanchard/will-it-scale
>>>>
>>>>
>>>> on test machine: 144 threads 4 sockets Intel(R) Xeon(R) Gold 5318H CPU @ 2.50GHz with 128G memory
>>>>
>>>> caused below changes (please refer to attached dmesg/kmsg for entire log/backtrace):
>>>>
>>>>
>>>>
>>>>
>>>> If you fix the issue, kindly add following tag
>>>> Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
>>>>
>>>>
>>>> [ 38.907274][ T868] nvme nvme0: pci function 0000:24:00.0
>>>> [ 38.924627][ T1103] scsi host0: ahci
>>>> 0m.
>>>> [ 38.948010][ T773] nvme nvme0: Identify Controller failed (16641)
>>>> [ 38.951220][ T1103] scsi host1: ahci
>>>> [ 38.954193][ T773] nvme nvme0: Removing after probe failure status: -5
>>>
>>> This is odd, looks like it's saying invalid opcode. Looking at the probe
>>> path, it's pretty standard and the command passed in is cleared already.
>>> So not quite sure why the patch would make a difference here. I'll
>>> poke at it.
>>
>> It's actually an Invalid Queue Identifier error (0x4101). That error
>> makes no sense for an Identify command, so it sounds like the controller
>> observed a different opcode than the driver intended to send, which
>> seems odd; I didn't observe any problems and I'm pretty sure I'm running
>> the same code. I'll take a second look as well.
>
> The git url that was used in this test points to commit:
>
> https://github.com/0day-ci/linux/commit/f9c499bbbf603389abad60d1931c16b2f96dee06
>
> And that commit has an extra memset in the REQ_OP_DRV_IN/OUT case, and
> it doesn't belong there. I don't see that memset in the upstream commit,
> Did the bot pick up the wrong patch?
Ah good catch, it's picking up a previous broken version. Good question on
why that might be, that's counter productive...
In any case, we can ignore it.
--
Jens Axboe
