[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <7c5beaf0-8fbc-d513-fffe-2140bbe5318f@suse.de>
Date: Thu, 14 Nov 2019 09:33:11 +0100
From: Thomas Zimmermann <tzimmermann@...e.de>
To: Gerd Hoffmann <kraxel@...hat.com>, drm@...hat.com
Cc: David Airlie <airlied@...ux.ie>,
"open list:DRM DRIVERS" <dri-devel@...ts.freedesktop.org>,
open list <linux-kernel@...r.kernel.org>,
Sean Paul <sean@...rly.run>
Subject: Re: [PATCH] drm/ttm: fix mmap refcounting
Am 13.11.19 um 14:56 schrieb Gerd Hoffmann:
> When mapping ttm objects via drm_gem_ttm_mmap() helper
> drm_gem_mmap_obj() will take an object reference. That gets
> never released due to ttm having its own reference counting.
>
> Fix that by dropping the gem object reference once the ttm mmap
> completed (and ttm refcount got bumped).
>
> For that to work properly the drm_gem_object_get() call in
> drm_gem_ttm_mmap() must be moved so it happens before calling
> obj->funcs->mmap(), otherwise the gem refcount would go down
> to zero.
>
> Fixes: 231927d939f0 ("drm/ttm: add drm_gem_ttm_mmap()")
> Signed-off-by: Gerd Hoffmann <kraxel@...hat.com>
Tested-by: Thomas Zimmermann <tzimmermann@...e.de>
> ---
> drivers/gpu/drm/drm_gem.c | 24 ++++++++++++++----------
> drivers/gpu/drm/drm_gem_ttm_helper.c | 13 ++++++++++++-
> 2 files changed, 26 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c
> index 2f2b889096b0..000fa4a1899f 100644
> --- a/drivers/gpu/drm/drm_gem.c
> +++ b/drivers/gpu/drm/drm_gem.c
> @@ -1105,21 +1105,33 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
> if (obj_size < vma->vm_end - vma->vm_start)
> return -EINVAL;
>
> + /* Take a ref for this mapping of the object, so that the fault
> + * handler can dereference the mmap offset's pointer to the object.
> + * This reference is cleaned up by the corresponding vm_close
> + * (which should happen whether the vma was created by this call, or
> + * by a vm_open due to mremap or partial unmap or whatever).
> + */
> + drm_gem_object_get(obj);
> +
> if (obj->funcs && obj->funcs->mmap) {
> /* Remove the fake offset */
> vma->vm_pgoff -= drm_vma_node_start(&obj->vma_node);
>
> ret = obj->funcs->mmap(obj, vma);
> - if (ret)
> + if (ret) {
> + drm_gem_object_put_unlocked(obj);
> return ret;
> + }
> WARN_ON(!(vma->vm_flags & VM_DONTEXPAND));
> } else {
> if (obj->funcs && obj->funcs->vm_ops)
> vma->vm_ops = obj->funcs->vm_ops;
> else if (dev->driver->gem_vm_ops)
> vma->vm_ops = dev->driver->gem_vm_ops;
> - else
> + else {
> + drm_gem_object_put_unlocked(obj);
> return -EINVAL;
> + }
>
> vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP;
> vma->vm_page_prot = pgprot_writecombine(vm_get_page_prot(vma->vm_flags));
> @@ -1128,14 +1140,6 @@ int drm_gem_mmap_obj(struct drm_gem_object *obj, unsigned long obj_size,
>
> vma->vm_private_data = obj;
>
> - /* Take a ref for this mapping of the object, so that the fault
> - * handler can dereference the mmap offset's pointer to the object.
> - * This reference is cleaned up by the corresponding vm_close
> - * (which should happen whether the vma was created by this call, or
> - * by a vm_open due to mremap or partial unmap or whatever).
> - */
> - drm_gem_object_get(obj);
> -
> return 0;
> }
> EXPORT_SYMBOL(drm_gem_mmap_obj);
> diff --git a/drivers/gpu/drm/drm_gem_ttm_helper.c b/drivers/gpu/drm/drm_gem_ttm_helper.c
> index 7412bfc5c05a..605a8a3da7f9 100644
> --- a/drivers/gpu/drm/drm_gem_ttm_helper.c
> +++ b/drivers/gpu/drm/drm_gem_ttm_helper.c
> @@ -64,8 +64,19 @@ int drm_gem_ttm_mmap(struct drm_gem_object *gem,
> struct vm_area_struct *vma)
> {
> struct ttm_buffer_object *bo = drm_gem_ttm_of_gem(gem);
> + int ret;
>
> - return ttm_bo_mmap_obj(vma, bo);
> + ret = ttm_bo_mmap_obj(vma, bo);
> + if (ret < 0)
> + return ret;
> +
> + /*
> + * ttm has its own object refcounting, so drop gem reference
> + * to avoid double accounting counting.
> + */
> + drm_gem_object_put_unlocked(gem);
> +
> + return 0;
> }
> EXPORT_SYMBOL(drm_gem_ttm_mmap);
>
>
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Maxfeldstr. 5, 90409 Nürnberg, Germany
(HRB 36809, AG Nürnberg)
Geschäftsführer: Felix Imendörffer
Download attachment "signature.asc" of type "application/pgp-signature" (489 bytes)
Powered by blists - more mailing lists