Re: [PATCH] drm/xe/pt: Reset prepare state in xe_pt_update_ops_init()

[Matthew Brost <matthew.brost@intel.com>]

On Mon, Jul 13, 2026 at 11:08:30PM +0000, Zongyao Bai wrote:
> xe_pt_update_ops_init() resets the per-prepare address range and
> deferred state, but current_op was left unchanged. If the same
> xe_vm_pgtable_update_ops is prepared again, new operations are appended
> after the stale current_op value instead of starting from ops[0]. This
> can make current_op exceed the number of operations allocated by
> xe_vma_ops_alloc(), leading to out-of-bounds access to pt_update_ops->ops.
> So, reset current_op when initializing the PT update ops.
> 
> Also reset needs_svm_lock and needs_invalidation. These flags are derived
> during the same prepare pass as current_op and are consumed by
> xe_pt_update_ops_run(); leaving them stale can select the SVM migrate ops
> or request TLB invalidation even when the current prepared operations do
> not require it.
> 

I'm not opposed to this patch as it makes retires with the same set ops
possible, perhaps makes the code safer but to be clear this isn't a
problem in the code base is it?

Assuming this isn't a problem what is the motivation here - future
proofing?

Matt

> Assisted-by: GitHub-Copilot:claude-sonnet-4.6
> Signed-off-by: Bai, Zongyao <zongyao.bai@intel.com>
> ---
>  drivers/gpu/drm/xe/xe_pt.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
> index e466f714bf86..598c6b2571e7 100644
> --- a/drivers/gpu/drm/xe/xe_pt.c
> +++ b/drivers/gpu/drm/xe/xe_pt.c
> @@ -2371,8 +2371,11 @@ static void
>  xe_pt_update_ops_init(struct xe_vm_pgtable_update_ops *pt_update_ops)
>  {
>  	init_llist_head(&pt_update_ops->deferred);
> +	pt_update_ops->current_op = 0;
>  	pt_update_ops->start = ~0x0ull;
>  	pt_update_ops->last = 0x0ull;
> +	pt_update_ops->needs_svm_lock = false;
> +	pt_update_ops->needs_invalidation = false;
>  	xe_page_reclaim_list_init(&pt_update_ops->prl);
>  }
>  
> -- 
> 2.43.0
>