On 05.11.26 01:01, Kavitha Daula wrote:
> From: Kavitha Daula <kavi@edera.dev>
> 
> Both callers of gnttab_unmap_refs_sync(), free_persistent_gnts() and
> xen_blkbk_unmap_purged_grants() in blkback, keep their struct
> gntab_unmap_queue_data on the stack. gnttab_unmap_refs_async() sets up
> the delayed work inside it with INIT_DELAYED_WORK(), so with
> CONFIG_DEBUG_OBJECTS_WORK and CONFIG_DEBUG_OBJECTS_TIMERS every
> persistent grant teardown warns twice, once for the work and once for
> its timer:
> 
>    [  104.366085] ODEBUG: object ffffc9004020fad8 is on stack ffffc9004020c000, but NOT annotated.
>    [  104.371551] WARNING: CPU: 1 PID: 48 at lib/debugobjects.c:672 lookup_object_or_alloc.part.0+0x12b/0x1f0
>    Call Trace:
>     <TASK>
>     __debug_object_init+0xfe/0x1a0
>     debug_object_init+0x38/0x50
>     __init_work+0x28/0x30
>     gnttab_unmap_refs_async+0x18/0xa0
>     gnttab_unmap_refs_sync+0x61/0x90
>     free_persistent_gnts+0x179/0x280
>     xen_blkbk_free_caches+0x16/0x30
>     xen_blkif_disconnect+0xfa/0x2d0
>     frontend_changed+0x90/0x680
>     xenbus_otherend_changed+0x80/0xb0
>     frontend_changed+0x14/0x20
>     xenwatch_thread+0x96/0x1a0
>     kthread+0x11d/0x250
>     ret_from_fork+0x29f/0x2d0
>     ret_from_fork_asm+0x1a/0x30
>     </TASK>
> 
> gnttab_unmap_refs_sync() waits for the unmap to complete before it
> returns, so the work never outlives the caller's frame. Initialise it
> with INIT_DELAYED_WORK_ONSTACK() when the item is on the stack and
> destroy it before returning. The function is exported, so an item that
> is not on the stack keeps INIT_DELAYED_WORK().
> 
> Fixes: b44166cd46e2 ("xen/grant: introduce func gnttab_unmap_refs_sync()")
> Signed-off-by: Kavitha Daula <kavi@edera.dev>
Reviewed-by: Juergen Gross <jgross@suse.com>


Juergen