[PATCH blktests] nvme/039: drain udev events before passthru error injection

[John Garry]

+ linux-nvme

On 30/06/2026 11:19, Mateusz Nowicki wrote:
> From: Mateusz Nowicki <Mateusz.Nowicki at posteo.net>
> 
> When run repeatedly, nvme/039 fails intermittently with a regular block
> read error instead of the expected passthru line, e.g.:
> 
>    nvme0n1: Read(0x2) @ LBA 0, 8 blocks, Invalid Command Opcode (sct 0x0 / sc 0x1) DNR

This is not the error which I was seeing, however this change seems to 
make the test pass reliably for the me. I have not debugged why the test 
was intermittently failing for me.

Please also note that I did trigger a kernel warn previously for this 
test and I posted a proposed fix in:

https://lore.kernel.org/linux-nvme/20260630102717.2671475-1-john.g.garry@oracle.com/T/#u

thanks

> 
> inject_write_fault_on_write() writes LBA 0 of the whole-disk device,
> whose release emits a change uevent that makes udev run blkid, reading
> LBA 0 asynchronously. The following passthru injection arms a one-shot
> fault (times=1); if the stray blkid read races in first it consumes the
> fault and is logged via nvme_log_error() (no cdw fields), so the expected
> passthru line is never emitted and the test fails.
> 
> Drain the pending uevent with udevadm settle before the passthru section.
> 
> Signed-off-by: Mateusz Nowicki <Mateusz.Nowicki at posteo.net>
> ---
>   tests/nvme/039 | 2 ++
>   1 file changed, 2 insertions(+)
> 
> diff --git a/tests/nvme/039 b/tests/nvme/039
> index 7ca48ec..8c70c5c 100755
> --- a/tests/nvme/039
> +++ b/tests/nvme/039
> @@ -206,6 +206,8 @@ test_device() {
>   	inject_invalid_status_on_read "${ns_dev}"
>   	inject_write_fault_on_write "${ns_dev}"
>   
> +	udevadm settle
> +
>   	if [ -e "$TEST_DEV_SYSFS/passthru_err_log_enabled" ]; then
>   		_nvme_passthru_logging_setup "${ns_dev}" "${ctrl_dev}"
>