NetBSD Problem Report #60558

From www@netbsd.org  Sat Aug  8 12:57:19 2026
Return-Path: <www@netbsd.org>
Received: from mail.netbsd.org (mail.netbsd.org [199.233.217.200])
	(using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)
	 key-exchange X25519 server-signature RSA-PSS (2048 bits)
	 client-signature RSA-PSS (2048 bits))
	(Client CN "mail.netbsd.org", Issuer "YR2" (not verified))
	by mollari.NetBSD.org (Postfix) with ESMTPS id B164A1A923B
	for <gnats-bugs@gnats.NetBSD.org>; Sat,  8 Aug 2026 12:57:18 +0000 (UTC)
Message-Id: <20260808125717.B8CD41A923C@mollari.NetBSD.org>
Date: Sat,  8 Aug 2026 12:57:17 +0000 (UTC)
From: campbell+netbsd@mumble.net
Reply-To: campbell+netbsd@mumble.net
To: gnats-bugs@NetBSD.org
Subject: nvmm(4): suspend/resume/detach issues
X-Send-Pr-Version: www-1.0
X-From4GNATS: "campbell+netbsd@mumble.net via gnats" <gnats-admin@NetBSD.org>

>Number:         60558
>Category:       kern
>Synopsis:       nvmm(4): suspend/resume/detach issues
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Aug 08 13:00:00 +0000 2026
>Originator:     Taylor R Campbell
>Release:        current, 11, 10
>Organization:
The NvmmBSD Suspension, Inc.
>Environment:
>Description:

	1. The logic to coordinate suspend/resume around nvmm ioctls
	   involves taking a global mutex to check whether we're
	   suspending.  This could be a performance bottleneck when
	   many VMs are running in parallel.

	2. nvmm_detach is racy, partly because of PR kern/60547:
	   fileops(9) can't be safely unloaded in a module.  Need to:

	   (a) mark that we're detaching,

	   (b) suspend nvmm ioctls and make sure nvmm_enter_sig fails
	       promptly when detaching,

	   (c) wait for all concurrent fileops to return, and then

	   (d) finally we can safely free all the resources.

>How-To-Repeat:

	code inspection

>Fix:

	1. pserialize(9)/localcount(9)

	2. (a) mark that we're detaching,
	   (b) suspend nvmm ioctls and make sure nvmm_enter_sig fails
	       promptly when detaching,
	   (c) wait for all concurrent fileops to return, and then
	   (d) finally we can safely free all the resources.

Home	PR Database Search