Re: [PATCH 6/8] drm/panthor: Introduce huge tmpfs mount point option

[Boris Brezillon]

On Tue, 30 Sep 2025 18:31:11 +0200
Loïc Molinari <loic.molinari@xxxxxxxxxxxxx> wrote:

> On 30/09/2025 12:34, Boris Brezillon wrote:
> > On Mon, 29 Sep 2025 22:03:14 +0200
> > Loïc Molinari <loic.molinari@xxxxxxxxxxxxx> wrote:
> >   
> >> diff --git a/drivers/gpu/drm/panthor/panthor_device.h b/drivers/gpu/drm/panthor/panthor_device.h
> >> index 4fc7cf2aeed5..54ca61567426 100644
> >> --- a/drivers/gpu/drm/panthor/panthor_device.h
> >> +++ b/drivers/gpu/drm/panthor/panthor_device.h
> >> @@ -135,6 +135,9 @@ struct panthor_device {
> >>   	/** @devfreq: Device frequency scaling management data. */
> >>   	struct panthor_devfreq *devfreq;
> >>   
> >> +	/** @huge_mnt: tmpfs mount point with Transparent Hugepage enabled. */
> >> +	struct vfsmount *huge_mnt;  
> > 
> > Now that we have a helper to create a huge mountpoint, wouldn't it
> > make sense to have this field in drm_device instead of having each
> > driver add a huge_mnt field to their <driver>_device object.  
> 
> Not sure this should be enforced for all DRM drivers since most of them 
> don't create separate huge mountpoints (only 4 for now including this 
> patchset) and I guess some maintainers might prefer to depend on the 
> sysfs interace to enable huge pages.

I'm not saying we should create the huge mountpoint by default, but if
this is a generic helper, it makes sense to also manage this mountpoint
internally. In the end, it'd be a small price to pay for drivers that
don't need it (the size of a pointer in the drm_device object), and
with this in place, driver wouldn't even have to call
drm_gem_shmem_huge_mnt_free() manually (can be automated with a
drmm_add_action_or_reset() calling kern_unmount() inside
drm_gem_shmem_huge_mnt_create()).

> 
> >> +
> >>   	/** @unplug: Device unplug related fields. */
> >>   	struct {
> >>   		/** @lock: Lock used to serialize unplug operations. */  
> >   
> 
> Loïc