[PATCH V2 1/1] nvme: introduce generic per-namespace chardev

Christoph Hellwig hch at lst.de
Wed Apr 7 16:40:00 BST 2021


On Thu, Apr 08, 2021 at 12:35:36AM +0900, Minwoo Im wrote:
> 1. During the `nvme list` command, controller ioctl for the ns_head has
>    been not coming out from the mutex_lock_killable(&nvme_subsystems_lock)
>    because it just gets the controller reference and return it without
>    unlocking it.  So the first change point of this patch is to unlock the
>    mutex right before the return.  But, Is this a real issue? because
>    this changes are not from this series though.....

Yeal issue, stupid rebase error when I stated to split and move a few
more pieces for better reuse from the chardev..

> 2. Can we have the check whether the ns_head has disk allocated or not
>    by getting `disk` pointer out of the #ifdef CONFIG_NVME_MULTIPATH?
>    If it's not allocated due to some reasons (e.g., !multipath, or CMIC
>    does not support multiple controllers, or some failures during the
>    allocations), disk will never be allocated.  So, I tried to pull the
>    `disk` pointer out of the #ifdef from the nvme_ns_head, but maybe
>    this is not what you have intended....  It would be great if you can
>    give some feedback on this.

This looks pretty sensible.  If you give me a signoff I'll commit this
as a separate prep patch.



More information about the Linux-nvme mailing list