[PATCH V3] nvme: enable char device per namespace

[Keith Busch]

On Tue, Dec 15, 2020 at 08:55:57PM +0100, javier at javigon.com wrote:
> From: Javier González <javier.gonz at samsung.com>
> 
> Create a char device per NVMe namespace. This char device is always
> initialized, independently of whether the features implemented by the
> device are supported by the kernel. User-space can therefore always
> issue IOCTLs to the NVMe driver using the char device.
> 
> The char device is presented as /dev/generic-nvmeXcYnZ. This naming
> scheme follows the convention of the hidden device (nvmeXcYnZ). Support
> for multipath will follow.
> 
> Christoph, Keith: Is this going in the right direction?

I think this is looking okay, though I'm getting some weird errors and
warnings during boot. The first one looks like the following (I will
look into it too, but I wanted to get a reply out sooner).

[    4.734143] sysfs: cannot create duplicate filename '/dev/char/242:1'
[    4.736359] CPU: 4 PID: 7 Comm: kworker/u16:0 Not tainted 5.10.0+ #172
[    4.740836] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
[    4.744228] Workqueue: nvme-wq nvme_scan_work [nvme_core]
[    4.745558] Call Trace:
[    4.746122]  dump_stack+0x6d/0x88
[    4.746834]  sysfs_warn_dup.cold+0x17/0x24
[    4.747656]  sysfs_do_create_link_sd.isra.0+0xb0/0xc0
[    4.747659]  device_add+0x604/0x7b0
[    4.749298]  cdev_device_add+0x46/0x70
[    4.753241]  ? cdev_init+0x51/0x60
[    4.753246]  nvme_alloc_ns+0x670/0x8a0 [nvme_core]
[    4.753249]  ? _cond_resched+0x15/0x30
[    4.753253]  nvme_validate_or_alloc_ns+0x99/0x190 [nvme_core]
[    4.753258]  nvme_scan_work+0x152/0x290 [nvme_core]
[    4.753261]  process_one_work+0x1ac/0x330
[    4.753262]  worker_thread+0x50/0x3a0
[    4.753264]  ? process_one_work+0x330/0x330
[    4.753265]  kthread+0xfb/0x130
[    4.753266]  ? kthread_park+0x90/0x90
[    4.753269]  ret_from_fork+0x1f/0x30
[    4.753272] CPU: 5 PID: 88 Comm: kworker/u16:1 Not tainted 5.10.0+ #172

> Keith: Regarding nvme-cli support, what do you think about reporting the
> char device as existing block devices? If this is OK with you, I will
> submit patches for the filters.

I'm not sure I understand what you mean about reporting these as
"existing block devices". Are you talking about what's shown in the
'nvme list' output?