Date: Tue, 16 Jun 2026 11:16:02 -0500 From: Matthew Grooms <mgrooms@shrew.net> To: virtualization@freebsd.org Subject: =?UTF-8?Q?Re=3A_**=5BHelp=5D_bhyve=3A_bootrom=5Falloc=3A_vm=5Fmmap?= =?UTF-8?Q?=5Fmapseg=3A_Invalid_argument_=E2=80=94_NVIDIA_passthrough_with_C?= =?UTF-8?Q?orvin=27s_branch_on_FreeBSD_15=2E0**?= Message-ID: <561d0b98-2183-4b0c-9843-38f897e65af5@shrew.net> In-Reply-To: <381ad530-f921-496d-b636-14ad28aa1e2d@shrew.net> References: <CA%2B1FSiiGdW4JCi=XA68Wtr1pDbMNAwTpwq9P8TixdDGRHSf%2B6g@mail.gmail.com> <bc674b3f2815204306480f88f71b48ff886c8a56.camel@FreeBSD.org> <1bbf28a7-7025-4d83-937d-72d8583048ca@shrew.net> <c55fa3d6d59284260d19064c25f66c4bcb385ef7.camel@FreeBSD.org> <381ad530-f921-496d-b636-14ad28aa1e2d@shrew.net>
index | next in thread | previous in thread | raw e-mail
If anyone else experiences this, disabling the frame buffer device appears to work around the issue for me. I can now shutdown and reboot the guest without a hang. Thanks again, -Matthew On 6/16/26 10:47, Matthew Grooms wrote: > Thanks Corvin. > > The passthrough feature appears to be working well. I appreciate the > time and effort you've put into this. Once passed, I can use the > device in a Debian 13 guest to execute cuda workloads. There do appear > to be a few issues: > > 1) The nvidia device is always claimed as the primary GPU by the guest. > 2) The guest hangs on shutdown and/or reboot. This only happens when > the GPU is passed through. > > The first issue prevents me from performing a device reset if the GPU > gets wedged in a CUDA operation. Reboot required. The second issue > prevents a clean reboot ( always a dirty FS at startup ). Any ideas > how to force UEFI to prefer the virtual frame buffer device? I've > tried everything I can think of but no luck so far. > > Thanks again, > > -Matthew > > On 6/8/26 01:39, Corvin Köhne wrote: >> On Fri, 2026-06-05 at 11:20 -0500, Matthew Grooms wrote: >>> On 6/5/26 01:34, Corvin Köhne wrote: >>>> On Thu, 2026-06-04 at 13:23 +0200, Mario Marietto wrote: >>>>> Hi everyone, >>>>> I'm trying to get NVIDIA GPU passthrough working with bhyve on >>>>> FreeBSD >>>>> 15.0- >>>>> RELEASE-p5, using Corvin Köhne's nvidia-wip branch: >>>>> https://github.com/Beckhoff/freebsd-src/tree/phab/corvink/15.0/nvidia-wip >>>>> The VM fails to start with the following error before the guest >>>>> even >>>>> boots: >>>>> bhyve: bootrom_alloc: vm_mmap_mapseg: Invalid argument >>>> NVIDIA GPU passthrough should work with stock 15.0, no patches >>>> required. >>> I was about to attempt this myself. Are all the required patches >>> present >>> in 14.4 as well? >>> >>> Thanks, >>> >>> -Matthew >>> >> Yes, the required patch is present in 14.4 as well. >> >> >home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?561d0b98-2183-4b0c-9843-38f897e65af5>