Okay so I was able to get it to start on auto with the following diff
(the xf86drm.h header has been established as already (sort of) merged,
but I hadn't included the "renderD" change in my own tree):

diff --git xf86drm.c xf86drm.c
index 2e76f0ea..6d29184f 100644
--- xf86drm.c
+++ xf86drm.c
@@ -3517,8 +3517,11 @@ static char *drmGetMinorNameForFD(int fd, int type)
 
     if (!dev_name)
         return NULL;
-
+#ifdef __OpenBSD__
+    n = snprintf(buf, sizeof(buf), dev_name, DRM_DIR_NAME, min+128);
+#else
     n = snprintf(buf, sizeof(buf), dev_name, DRM_DIR_NAME, min);
+#endif
     if (n == -1 || n >= sizeof(buf))
         return NULL;
 
@@ -4585,6 +4588,7 @@ drm_public int drmGetDeviceFromDevId(dev_t find_rdev, uint32_t flags, drmDeviceP
     const char      *dev_name;
     int              node_type, subsystem_type;
     int              maj, min, n, ret;
+    struct stat sbuf;
 
     if (device == NULL)
         return -EINVAL;
diff --git xf86drm.h xf86drm.h
index 2cb14156..6254d912 100644
--- xf86drm.h
+++ xf86drm.h
@@ -77,10 +77,10 @@ extern "C" {
 #define DRM_DEV_MODE	 (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP)
 
 #ifdef __OpenBSD__
-#define DRM_DIR_NAME  "/dev"
-#define DRM_PRIMARY_MINOR_NAME  "drm"
+#define DRM_DIR_NAME  "/dev/dri"
+#define DRM_PRIMARY_MINOR_NAME  "card"
 #define DRM_CONTROL_MINOR_NAME  "drmC" /* deprecated */
-#define DRM_RENDER_MINOR_NAME   "drmR"
+#define DRM_RENDER_MINOR_NAME   "renderD"
 #else
 #define DRM_DIR_NAME  "/dev/dri"
 #define DRM_PRIMARY_MINOR_NAME  "card"


On Thu, Jan 04, 2024 at 09:20:18AM -0700, Chris Waddey wrote:
> I forgot that I needed to add this diff. There is an "#ifdef __OpenBSD__" just
> above this and a comment that looks a bit dated in the latest drm code.
> 
> diff --git xf86drm.c xf86drm.c
> index 2e76f0ea..7c4f5cf0 100644
> --- xf86drm.c
> +++ xf86drm.c
> @@ -4585,6 +4585,7 @@ drm_public int drmGetDeviceFromDevId(dev_t find_rdev, uint32_t flags, drmDeviceP
>      const char      *dev_name;
>      int              node_type, subsystem_type;
>      int              maj, min, n, ret;
> +    struct stat sbuf;
> 
>      if (device == NULL)
>          return -EINVAL;
> 
> On Thu, Jan 04, 2024 at 11:07:56AM +0100, Matthieu Herrb wrote:
> > On Wed, Jan 03, 2024 at 07:52:56PM -0700, Chris Waddey wrote:
> > > Hello.
> > 
> > Hi,
> > 
> > > I was able to get the latest version of sway working on my box after
> > > getting the latest version of wlroots to work, which was after applying
> > > the following patch to the latest version of libdrm (source obtained
> > > from https://gitlab.freedesktop.org/mesa/drm.git). I don't know if we
> > > have some patches somewhere else that handle this, but I figured I
> > > would send it your way in case it would save some effort.
> > 
> > Does this include the removal of the wlr_drm protocol which was
> > committed to sway/wlroots a few hours ago ?
> 
> So I just did a fresh pull/build/install of libdrm, wlroots, and
> sway. Still working with the caveats below.
>  
> > I wasn't sure if OpenBSD libdrm / mesa supports linux-dmabuf which is
> > supposed to replace it.
> > 
> > If it works with linux-dmabuf, that's great news. I don't think there
> > are any other significant changes since the released versions packaged
> > in ports that could impact OpenBSD.
> > 
> > > I also had to do
> > > 
> > > doas chmod $USER:$USER /dev/dri/* && env WLR_RENDERER=pixman
> > > startsway.sh
> > 
> > Normally if you log in on the text console where you want to run sway,
> > the /dev/dri/* owndership  is taken care of by fbtab(5).
> 
> Yeah, I don't know why I included the chown bit. Off-topic I think.
> 
> > I think setting WLR_RENDERER to pixman means that you don't use the
> > new linux-dmabuf protocol. What happens if you leave that variable
> > undefined (defaults to 'auto') ?
> 
> It only works with WLR_RENDERER=pixman both before and after the
> recompile this morning. I've tried auto (which tries gles2) as well as
> vulkan. Sorry if this is just noise.
> 
> > > to get everything just right. And of course I applied the patches I
> > > found in ports to sway and wlroots, but nothing else that I can
> > > think of.
> > 
> > This diff is present in the /usr/xenocara/lib/libdrm tree. (not
> > exactly the same though: the whole #ifdef __OpenBSD__ part can be
> > removed now).
> 
> I think I see this now. I figured it lived somewhere in there.
> 
> > > 
> > > ~ 0$ pkg-config --modversion libdrm
> > > 2.4.119
> > > ~ 0$ pkg-config --modversion wlroots
> > > 0.18.0-dev
> > > ~ 0$ sway --version
> > > sway version 1.9-dev-52eaa5be (Jan  3 2024, branch 'master')
> > > 
> > > diff --git xf86drm.h xf86drm.h
> > > index 2cb14156..c91f157b 100644
> > > --- xf86drm.h
> > > +++ xf86drm.h
> > > @@ -77,8 +77,8 @@ extern "C" {
> > >  #define DRM_DEV_MODE	 (S_IRUSR|S_IWUSR|S_IRGRP|S_IWGRP)
> > >  
> > >  #ifdef __OpenBSD__
> > > -#define DRM_DIR_NAME  "/dev"
> > > -#define DRM_PRIMARY_MINOR_NAME  "drm"
> > > +#define DRM_DIR_NAME  "/dev/dri"
> > > +#define DRM_PRIMARY_MINOR_NAME  "card"
> > >  #define DRM_CONTROL_MINOR_NAME  "drmC" /* deprecated */
> > >  #define DRM_RENDER_MINOR_NAME   "drmR"
> > >  #else
> > > 
> > 
>