Tested FreeBSD 15.0 64bit ALPHA3 ISO Image installation on vSphere, and found that the VM needs reboot after hot adding a NVMe hard disk to an existing NVMe controller or hot removing a NVMe hard disk from a NVMe controller. But for FreeBSD 14.x like FreeBSD 14.3, it doesn't need to reboot for recognizing the newly added NVMe hard disk like by lsblk command. # lsblk DEVICE MAJ:MIN SIZE TYPE LABEL MOUNT da0 0:86 20G GPT - - da0p1 0:87 512K freebsd-boot gpt/gptboot0 - <FREE> -:- 492K - - - da0p2 0:88 2.0G freebsd-swap gpt/freebsd-swap SWAP da0p3 0:89 18G freebsd-zfs gpt/freebsd-zfs <ZFS> <FREE> -:- 1.0M - - - nda0 0:83 1.0G - - - Would you please take a look at this issue on FreeBSD 15.0? Thanks! Yanhui
Created attachment 264613 [details] nvme crash on resume from suspend.
(In reply to Fabio Comolli from comment #1) sorry,the picture was meant to be attached to a different bug.
Hi, Would you please take a look? It still happens on FreeBSD 15.0 BETA5. Thanks! Yanhui
Still happens on FreeBSD 15.0 RC1.
It's not clear to me this is an nvme related issue. The nvme controller isn't responding at all after the resume. The root cause for that is needed. If there's a hot remove of root, that's never going to work. So how do I recreate this with software I might have access to? I don't have vsphere. So I'm a bit confused. The description on the attachment says something about suspend/resume, but the other talks about a hot add. I am aware that changing the size of an nvme namespace won't work in 15.0, but will shortly in -current. But that's not adding a namespace to an existing controller. I do see a change that might cause that to fail in 15 but not 14. So I think I need a few more details from this shorthand to really know what's going on.
If you're talking about my attachment please note that it's not related to this bug, it was a mistake on my part. I don't seem to be able to delete it.
(In reply to Fabio Comolli from comment #6) The attachment is related to https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=290265
I think there's a small chance that commit 27481c268916b0790c7ad16202a5b012625ce1a8 Author: Warner Losh <imp@FreeBSD.org> Date: Tue Nov 18 13:07:11 2025 -0700 nvme: Fix backwards sense of error condition b21e67875bf0c tested for the good condition, not the error condition, so we'd never do anything else in this function. This was causing certain logging not to happen, and also prevented forthcoming namespace size change code from working as well. Fixes: b21e67875bf0c Sponsored by: Netflix will fix this specific issue. But it may be too late to get into 15.0. I also think that the rest of the patches from there through commit 4640f5008922c5b189d2f7b63edf73300277e6df (HEAD -> main, freebsd/main, freebsd/HEAD) Author: Wanpeng Qian <wanpengqian@gmail.com> Date: Tue Nov 18 10:24:13 2025 -0700 nvme_sim: signal namespace depature Signal when the namespace is gone so we can tear down the disk when a nvme drive is removed. Reviewed by: imp Differential Revision: https://reviews.freebsd.org/D33032 have an excellent chance of making life a lot better for you in general.
(In reply to Warner Losh from comment #8) Thanks Warner! We have finished the testing for FreeBSD 15.0, and will note this known issue for the next release of FreeBSD 15.1.