Re: [PATCH] drm/msm: Fix GEM free for imported dma-bufs

[Stephan Gerhold <stephan.gerhold@linaro.org>]

On Tue, Sep 23, 2025 at 07:04:40AM -0700, Rob Clark wrote:
> Imported dma-bufs also have obj->resv != &obj->_resv.  So we should
> check both this condition in addition to flags for handling the
> _NO_SHARE case.
> 
> Fixes this splat that was reported with IRIS video playback:
> 
>     ------------[ cut here ]------------
>     WARNING: CPU: 3 PID: 2040 at drivers/gpu/drm/msm/msm_gem.c:1127 msm_gem_free_object+0x1f8/0x264 [msm]
>     CPU: 3 UID: 1000 PID: 2040 Comm: .gnome-shell-wr Not tainted 6.17.0-rc7 #1 PREEMPT
>     pstate: 81400005 (Nzcv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
>     pc : msm_gem_free_object+0x1f8/0x264 [msm]
>     lr : msm_gem_free_object+0x138/0x264 [msm]
>     sp : ffff800092a1bb30
>     x29: ffff800092a1bb80 x28: ffff800092a1bce8 x27: ffffbc702dbdbe08
>     x26: 0000000000000008 x25: 0000000000000009 x24: 00000000000000a6
>     x23: ffff00083c72f850 x22: ffff00083c72f868 x21: ffff00087e69f200
>     x20: ffff00087e69f330 x19: ffff00084d157ae0 x18: 0000000000000000
>     x17: 0000000000000000 x16: ffffbc704bd46b80 x15: 0000ffffd0959540
>     x14: 0000000000000000 x13: 0000000000000000 x12: 0000000000000000
>     x11: ffffbc702e6cdb48 x10: 0000000000000000 x9 : 000000000000003f
>     x8 : ffff800092a1ba90 x7 : 0000000000000000 x6 : 0000000000000020
>     x5 : ffffbc704bd46c40 x4 : fffffdffe102cf60 x3 : 0000000000400032
>     x2 : 0000000000020000 x1 : ffff00087e6978e8 x0 : ffff00087e6977e8
>     Call trace:
>      msm_gem_free_object+0x1f8/0x264 [msm] (P)
>      drm_gem_object_free+0x1c/0x30 [drm]
>      drm_gem_object_handle_put_unlocked+0x138/0x150 [drm]
>      drm_gem_object_release_handle+0x5c/0xcc [drm]
>      drm_gem_handle_delete+0x68/0xbc [drm]
>      drm_gem_close_ioctl+0x34/0x40 [drm]
>      drm_ioctl_kernel+0xc0/0x130 [drm]
>      drm_ioctl+0x360/0x4e0 [drm]
>      __arm64_sys_ioctl+0xac/0x104
>      invoke_syscall+0x48/0x104
>      el0_svc_common.constprop.0+0x40/0xe0
>      do_el0_svc+0x1c/0x28
>      el0_svc+0x34/0xec
>      el0t_64_sync_handler+0xa0/0xe4
>      el0t_64_sync+0x198/0x19c
>     ---[ end trace 0000000000000000 ]---
>     ------------[ cut here ]------------
> 
> Reported-by: Stephan Gerhold <stephan.gerhold@linaro.org>

Tested-by: Stephan Gerhold <stephan.gerhold@linaro.org>

Perhaps also add

Closes: https://lore.kernel.org/r/aNF6N8u1VIFSTaRM@linaro.org/

when applying.

Thanks!
Stephan

> Fixes: de651b6e040b ("drm/msm: Fix refcnt underflow in error path")
> Signed-off-by: Rob Clark <robin.clark@oss.qualcomm.com>
> ---
>  drivers/gpu/drm/msm/msm_gem.c | 10 +++++++---
>  1 file changed, 7 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpu/drm/msm/msm_gem.c b/drivers/gpu/drm/msm/msm_gem.c
> index 9f0f5b77f1bd..3aea9b493375 100644
> --- a/drivers/gpu/drm/msm/msm_gem.c
> +++ b/drivers/gpu/drm/msm/msm_gem.c
> @@ -1121,12 +1121,16 @@ static void msm_gem_free_object(struct drm_gem_object *obj)
>  		put_pages(obj);
>  	}
>  
> -	if (obj->resv != &obj->_resv) {
> +	/*
> +	 * In error paths, we could end up here before msm_gem_new_handle()
> +	 * has changed obj->resv to point to the shared resv.  In this case,
> +	 * we don't want to drop a ref to the shared r_obj that we haven't
> +	 * taken yet.
> +	 */
> +	if ((msm_obj->flags & MSM_BO_NO_SHARE) && (obj->resv != &obj->_resv)) {
>  		struct drm_gem_object *r_obj =
>  			container_of(obj->resv, struct drm_gem_object, _resv);
>  
> -		WARN_ON(!(msm_obj->flags & MSM_BO_NO_SHARE));
> -
>  		/* Drop reference we hold to shared resv obj: */
>  		drm_gem_object_put(r_obj);
>  	}
> -- 
> 2.51.0
>