drm/i915: Prevent unbounded wm results in g4x_compute_wm()
authorChris Wilson <chris@chris-wilson.co.uk>
Tue, 7 Nov 2017 14:03:38 +0000 (14:03 +0000)
committerChris Wilson <chris@chris-wilson.co.uk>
Tue, 7 Nov 2017 17:50:32 +0000 (17:50 +0000)
Smatch warns of

drivers/gpu/drm/i915/intel_pm.c:1161 g4x_compute_wm() warn: signedness bug returning '(-33554430)'

which is a result of it believing that wm may be INT_MAX following
g4x_tlb_miss_wa(). Just declaring g4x_tlb_miss_wa() as returning an
unsigned integer is not sufficient, we need to tell smatch that wm itself
is unsigned for it to not worry. So mark up the locals we expect to be
non-negative, and so silence smatch.

v2: Mark up vlv_compute_wm_level() as unsigned similarly.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> #v1
Link: https://patchwork.freedesktop.org/patch/msgid/20171107140338.13748-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/intel_pm.c

index 07118c0..46440e2 100644 (file)
@@ -928,7 +928,7 @@ static void pineview_update_wm(struct intel_crtc *unused_crtc)
  *  and the size of 8 whole lines. This adjustment is always performed
  *  in the actual pixel depth regardless of whether FBC is enabled or not."
  */
-static int g4x_tlb_miss_wa(int fifo_size, int width, int cpp)
+static unsigned int g4x_tlb_miss_wa(int fifo_size, int width, int cpp)
 {
        int tlb_miss = fifo_size * 64 - width * cpp * 8;
 
@@ -1105,8 +1105,8 @@ static uint16_t g4x_compute_wm(const struct intel_crtc_state *crtc_state,
        struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
        const struct drm_display_mode *adjusted_mode =
                &crtc_state->base.adjusted_mode;
-       int clock, htotal, cpp, width, wm;
-       int latency = dev_priv->wm.pri_latency[level] * 10;
+       unsigned int latency = dev_priv->wm.pri_latency[level] * 10;
+       unsigned int clock, htotal, cpp, width, wm;
 
        if (latency == 0)
                return USHRT_MAX;
@@ -1145,7 +1145,7 @@ static uint16_t g4x_compute_wm(const struct intel_crtc_state *crtc_state,
                   level == G4X_WM_LEVEL_NORMAL) {
                wm = intel_wm_method1(clock, cpp, latency);
        } else {
-               int small, large;
+               unsigned int small, large;
 
                small = intel_wm_method1(clock, cpp, latency);
                large = intel_wm_method2(clock, htotal, width, cpp, latency);
@@ -1158,7 +1158,7 @@ static uint16_t g4x_compute_wm(const struct intel_crtc_state *crtc_state,
 
        wm = DIV_ROUND_UP(wm, 64) + 2;
 
-       return min_t(int, wm, USHRT_MAX);
+       return min_t(unsigned int, wm, USHRT_MAX);
 }
 
 static bool g4x_raw_plane_wm_set(struct intel_crtc_state *crtc_state,
@@ -1602,7 +1602,7 @@ static uint16_t vlv_compute_wm_level(const struct intel_crtc_state *crtc_state,
        struct drm_i915_private *dev_priv = to_i915(plane->base.dev);
        const struct drm_display_mode *adjusted_mode =
                &crtc_state->base.adjusted_mode;
-       int clock, htotal, cpp, width, wm;
+       unsigned int clock, htotal, cpp, width, wm;
 
        if (dev_priv->wm.pri_latency[level] == 0)
                return USHRT_MAX;
@@ -1628,7 +1628,7 @@ static uint16_t vlv_compute_wm_level(const struct intel_crtc_state *crtc_state,
                                    dev_priv->wm.pri_latency[level] * 10);
        }
 
-       return min_t(int, wm, USHRT_MAX);
+       return min_t(unsigned int, wm, USHRT_MAX);
 }
 
 static bool vlv_need_sprite0_fifo_workaround(unsigned int active_planes)