Hi, On 11/09/2024 06:39, Upadhyay, Tejas wrote:
-----Original Message----- From: Auld, Matthew <matthew.auld@xxxxxxxxx> Sent: Tuesday, September 10, 2024 6:42 PM To: intel-xe@xxxxxxxxxxxxxxxxxxxxx Cc: Ghimiray, Himal Prasad <himal.prasad.ghimiray@xxxxxxxxx>; Upadhyay, Tejas <tejas.upadhyay@xxxxxxxxx>; Thomas Hellström <thomas.hellstrom@xxxxxxxxxxxxxxx>; stable@xxxxxxxxxxxxxxx Subject: [PATCH 2/4] drm/xe/client: add missing bo locking in show_meminfo() bo_meminfo() wants to inspect bo state like tt and the ttm resource, however this state can change at any point leading to stuff like NPD and UAF, if the bo lock is not held. Grab the bo lock when calling bo_meminfo(), ensuring we drop any spinlocks first. In the case of object_idr we now also need to hold a ref. Fixes: 0845233388f8 ("drm/xe: Implement fdinfo memory stats printing") Signed-off-by: Matthew Auld <matthew.auld@xxxxxxxxx> Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@xxxxxxxxx> Cc: Tejas Upadhyay <tejas.upadhyay@xxxxxxxxx> Cc: "Thomas Hellström" <thomas.hellstrom@xxxxxxxxxxxxxxx> Cc: <stable@xxxxxxxxxxxxxxx> # v6.8+ --- drivers/gpu/drm/xe/xe_drm_client.c | 37 +++++++++++++++++++++++++++--- 1 file changed, 34 insertions(+), 3 deletions(-) diff --git a/drivers/gpu/drm/xe/xe_drm_client.c b/drivers/gpu/drm/xe/xe_drm_client.c index badfa045ead8..3cca741c500c 100644 --- a/drivers/gpu/drm/xe/xe_drm_client.c +++ b/drivers/gpu/drm/xe/xe_drm_client.c @@ -10,6 +10,7 @@ #include <linux/slab.h> #include <linux/types.h> +#include "xe_assert.h" #include "xe_bo.h" #include "xe_bo_types.h" #include "xe_device_types.h" @@ -151,10 +152,13 @@ void xe_drm_client_add_bo(struct xe_drm_client *client, */ void xe_drm_client_remove_bo(struct xe_bo *bo) { + struct xe_device *xe = ttm_to_xe_device(bo->ttm.bdev); struct xe_drm_client *client = bo->client; + xe_assert(xe, !kref_read(&bo->ttm.base.refcount)); + spin_lock(&client->bos_lock); - list_del(&bo->client_link); + list_del_init(&bo->client_link); spin_unlock(&client->bos_lock); xe_drm_client_put(client); @@ -207,7 +211,20 @@ static void show_meminfo(struct drm_printer *p, struct drm_file *file) idr_for_each_entry(&file->object_idr, obj, id) { struct xe_bo *bo = gem_to_xe_bo(obj); - bo_meminfo(bo, stats); + if (dma_resv_trylock(bo->ttm.base.resv)) { + bo_meminfo(bo, stats); + xe_bo_unlock(bo); + } else { + xe_bo_get(bo); + spin_unlock(&file->table_lock); + + xe_bo_lock(bo, false); + bo_meminfo(bo, stats); + xe_bo_unlock(bo); + + xe_bo_put(bo); + spin_lock(&file->table_lock); + } } spin_unlock(&file->table_lock); @@ -217,7 +234,21 @@ static void show_meminfo(struct drm_printer *p, struct drm_file *file) if (!kref_get_unless_zero(&bo->ttm.base.refcount)) continue;While we have ref to BO, why would it need lock here, can you please explain if I am missing something. I though BO cant be deleted till will hold ref?
The ref is just about protecting the lifetime of the bo, however the internal bo state in particular the ttm stuff, is all protected by holding the dma-resv bo lock.
For example the bo can be moved/evicted around at will and the object state changes with it, but that should be done only when also holding the bo lock. If we are holding the bo lock here then the object state should be stable, making it safe to inspect stuff like bo->ttm.ttm and bo->ttm.resource. As an example, if you look at ttm_bo_move_null() and imagine xe_bo_has_pages() racing with that, then NPD or UAF is possible.
Thanks, Tejas- bo_meminfo(bo, stats); + if (dma_resv_trylock(bo->ttm.base.resv)) { + bo_meminfo(bo, stats); + xe_bo_unlock(bo); + } else { + spin_unlock(&client->bos_lock); + + xe_bo_lock(bo, false); + bo_meminfo(bo, stats); + xe_bo_unlock(bo); + + spin_lock(&client->bos_lock); + /* The bo ref will prevent this bo from being removed from the list */ + xe_assert(xef->xe, !list_empty(&bo->client_link)); + } + xe_bo_put_deferred(bo, &deferred); } spin_unlock(&client->bos_lock); -- 2.46.0
