Re: [PATCH linux-next] i915: intel_set_mode: Reduce stack allocation from 500 bytes to 2 pointers

From: Jani Nikula <jani.nikula@linux.intel.com>
To: linux-kernel@vger.kernel.org
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>,
	Tim Gardner <tim.gardner@canonical.com>,
	dri-devel@lists.freedesktop.org
Subject: Re: [PATCH linux-next] i915: intel_set_mode: Reduce stack allocation from 500 bytes to 2 pointers
Date: Fri, 07 Dec 2012 13:47:46 +0200	[thread overview]
Message-ID: <87zk1qm6cd.fsf@intel.com> (raw)
In-Reply-To: <1354815755-83271-1-git-send-email-tim.gardner@canonical.com>

On Thu, 06 Dec 2012, Tim Gardner <tim.gardner@canonical.com> wrote:
> smatch warning:
>
> drivers/gpu/drm/i915/intel_display.c:7019 intel_set_mode() warn: function puts
> 500 bytes on stack
>
> Refactor so that saved_mode and saved_hwmode are dynamically allocated as opposed
> to being automatic variables. 500 bytes seems like it could run the potential for blowing
> the kernel stack, especially if struct drm_display_mode ever grows significantly.

Hi Tim, seems reasonable, but please find a few comments inline.

BR,
Jani.

>
> Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
> Cc: David Airlie <airlied@linux.ie>
> Cc: dri-devel@lists.freedesktop.org
> Signed-off-by: Tim Gardner <tim.gardner@canonical.com>
> ---
>  drivers/gpu/drm/i915/intel_display.c |   18 +++++++++++++-----
>  1 file changed, 13 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index de51489..f8754d6 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -7739,11 +7739,18 @@ bool intel_set_mode(struct drm_crtc *crtc,
>  {
>  	struct drm_device *dev = crtc->dev;
>  	drm_i915_private_t *dev_priv = dev->dev_private;
> -	struct drm_display_mode *adjusted_mode, saved_mode, saved_hwmode;
> +	struct drm_display_mode *adjusted_mode, *saved_mode, *saved_hwmode;
>  	struct intel_crtc *intel_crtc;
>  	unsigned disable_pipes, prepare_pipes, modeset_pipes;
>  	bool ret = true;
>  
> +	saved_mode = kmalloc(2*sizeof(*saved_mode), GFP_KERNEL);

Style nitpick, missing spaces around the first *.

> +	if (!saved_mode) {
> +		pr_err("i915: Could not allocate saved display mode.\n");

Please use DRM_ERROR().

> +		return false;
> +	}
> +	saved_hwmode = saved_mode + 1;
> +
>  	intel_modeset_affected_pipes(crtc, &modeset_pipes,
>  				     &prepare_pipes, &disable_pipes);
>  
> @@ -7753,8 +7760,8 @@ bool intel_set_mode(struct drm_crtc *crtc,
>  	for_each_intel_crtc_masked(dev, disable_pipes, intel_crtc)
>  		intel_crtc_disable(&intel_crtc->base);
>  
> -	saved_hwmode = crtc->hwmode;
> -	saved_mode = crtc->mode;
> +	*saved_hwmode = crtc->hwmode;
> +	*saved_mode = crtc->mode;
>  
>  	/* Hack: Because we don't (yet) support global modeset on multiple
>  	 * crtcs, we don't keep track of the new mode for more than one crtc.

You are missing a kfree in the adjusted_mode error path after this hunk.

> @@ -7817,12 +7824,13 @@ bool intel_set_mode(struct drm_crtc *crtc,
>  done:
>  	drm_mode_destroy(dev, adjusted_mode);
>  	if (!ret && crtc->enabled) {
> -		crtc->hwmode = saved_hwmode;
> -		crtc->mode = saved_mode;
> +		crtc->hwmode = *saved_hwmode;
> +		crtc->mode = *saved_mode;
>  	} else {
>  		intel_modeset_check_state(dev);
>  	}
>  
> +	kfree(saved_mode);
>  	return ret;
>  }
>  
> -- 
> 1.7.9.5
>
> _______________________________________________
> dri-devel mailing list
> dri-devel@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel