Re: [PATCH] staging: media: atomisp: fix map and vmap leaks in stat buffer allocation

[Andy Shevchenko]

On Thu, Apr 16, 2026 at 03:27:31PM +0800, Huihui Huang wrote:
> Our code analyzer reported memory leaks in
> drivers/staging/media/atomisp/pci/atomisp_compat_css20.c.
> 
> In atomisp_css_allocate_stat_buffers(), s3a_map is allocated by
> ia_css_isp_3a_statistics_map_allocate() and its backing memory is
> mapped via hmm_vmap(). When dis_buf allocation fails, the error path
> frees s3a_data but does not unmap or free s3a_map. Similarly, when
> md_buf allocation fails, neither s3a_map nor dvs_map (and their hmm
> vmaps) are freed.

> My patch adds the missing hmm_vunmap() and map free calls on both
> error paths, matching the cleanup order used in
> atomisp_css_free_3a_buffer() and atomisp_css_free_dis_buffer().

Imperative mood.

...

>  			dev_err(isp->dev, "dvs buf allocation failed.\n");
> -			if (s3a_buf)
> +			if (s3a_buf) {
> +				hmm_vunmap(s3a_buf->s3a_data->data_ptr);
> +				ia_css_isp_3a_statistics_map_free(s3a_buf->s3a_map);
> +				s3a_buf->s3a_map = NULL;

Are these NULLifications needed? It sounds like it tries to paper over some
potential UAF cases. Is there any possibility to access s3a_map at this point?

>  				ia_css_isp_3a_statistics_free(s3a_buf->s3a_data);
> +			}

-- 
With Best Regards,
Andy Shevchenko