Re: [PATCH v2] drm/panthor: Fix potential memleak of vma structure

[Boris Brezillon <boris.brezillon@collabora.com>]

On Mon, 20 Oct 2025 21:02:43 +0100
Akash Goel <akash.goel@arm.com> wrote:

> This commit addresses a memleak issue of panthor_vma (or drm_gpuva)
> structure in Panthor driver, that can happen if the GPU page table
> update operation to map the pages fail.
> The issue is very unlikely to occur in practice.
> 
> v2: Add panthor_vm_op_ctx_return_vma() helper (Boris)
> 
> Fixes: 647810ec2476 ("drm/panthor: Add the MMU/VM logical block")
> Signed-off-by: Akash Goel <akash.goel@arm.com>

Just one minor thing below, but the patch is

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>

regardless.

> ---
>  drivers/gpu/drm/panthor/panthor_mmu.c | 16 +++++++++++++++-
>  1 file changed, 15 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_mmu.c b/drivers/gpu/drm/panthor/panthor_mmu.c
> index 6dec4354e378..63af8ee89b08 100644
> --- a/drivers/gpu/drm/panthor/panthor_mmu.c
> +++ b/drivers/gpu/drm/panthor/panthor_mmu.c
> @@ -1146,6 +1146,18 @@ static void panthor_vm_cleanup_op_ctx(struct panthor_vm_op_ctx *op_ctx,
>  	}
>  }
>  
> +static void
> +panthor_vm_op_ctx_return_vma(struct panthor_vm_op_ctx *op_ctx,
> +			     struct panthor_vma *vma)
> +{
> +	for (u32 i = 0; i < ARRAY_SIZE(op_ctx->preallocated_vmas); i++) {
> +		if (!op_ctx->preallocated_vmas[i]) {
> +			op_ctx->preallocated_vmas[i] = vma;
> +			return;
> +		}
> +	}

Maybe add a WARN_ON_ONCE() here.

> +}
> +
>  static struct panthor_vma *
>  panthor_vm_op_ctx_get_vma(struct panthor_vm_op_ctx *op_ctx)
>  {
> @@ -2081,8 +2093,10 @@ static int panthor_gpuva_sm_step_map(struct drm_gpuva_op *op, void *priv)
>  	ret = panthor_vm_map_pages(vm, op->map.va.addr, flags_to_prot(vma->flags),
>  				   op_ctx->map.sgt, op->map.gem.offset,
>  				   op->map.va.range);
> -	if (ret)
> +	if (ret) {
> +		panthor_vm_op_ctx_return_vma(op_ctx, vma);
>  		return ret;
> +	}
>  
>  	/* Ref owned by the mapping now, clear the obj field so we don't release the
>  	 * pinning/obj ref behind GPUVA's back.