drm/i915: Fix unsigned overflow when calculating total data rate, v2.
authorMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Mon, 22 Oct 2018 10:20:00 +0000 (12:20 +0200)
committerMaarten Lankhorst <maarten.lankhorst@linux.intel.com>
Wed, 24 Oct 2018 08:28:44 +0000 (10:28 +0200)
On gen11, we can definitely smash the 32-bits barrier with just a
when we enable all planes in the next patch.

Changes since v1:
- Use div64_u64 (ickle).

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20181022102000.30255-1-maarten.lankhorst@linux.intel.com
drivers/gpu/drm/i915/intel_pm.c

index 67a4d0735291a235a99f3ead443eada6b99f48cc..f42b8c319046492771e27658186bd994c8aaf0f4 100644 (file)
@@ -3784,7 +3784,7 @@ bool intel_can_enable_sagv(struct drm_atomic_state *state)
 
 static u16 intel_get_ddb_size(struct drm_i915_private *dev_priv,
                              const struct intel_crtc_state *cstate,
-                             const unsigned int total_data_rate,
+                             const u64 total_data_rate,
                              const int num_active,
                              struct skl_ddb_allocation *ddb)
 {
@@ -3798,12 +3798,12 @@ static u16 intel_get_ddb_size(struct drm_i915_private *dev_priv,
                return ddb_size - 4; /* 4 blocks for bypass path allocation */
 
        adjusted_mode = &cstate->base.adjusted_mode;
-       total_data_bw = (u64)total_data_rate * drm_mode_vrefresh(adjusted_mode);
+       total_data_bw = total_data_rate * drm_mode_vrefresh(adjusted_mode);
 
        /*
         * 12GB/s is maximum BW supported by single DBuf slice.
         */
-       if (total_data_bw >= GBps(12) || num_active > 1) {
+       if (num_active > 1 || total_data_bw >= GBps(12)) {
                ddb->enabled_slices = 2;
        } else {
                ddb->enabled_slices = 1;
@@ -3816,7 +3816,7 @@ static u16 intel_get_ddb_size(struct drm_i915_private *dev_priv,
 static void
 skl_ddb_get_pipe_allocation_limits(struct drm_device *dev,
                                   const struct intel_crtc_state *cstate,
-                                  const unsigned int total_data_rate,
+                                  const u64 total_data_rate,
                                   struct skl_ddb_allocation *ddb,
                                   struct skl_ddb_entry *alloc, /* out */
                                   int *num_active /* out */)
@@ -4139,7 +4139,7 @@ int skl_check_pipe_max_pixel_rate(struct intel_crtc *intel_crtc,
        return 0;
 }
 
-static unsigned int
+static u64
 skl_plane_relative_data_rate(const struct intel_crtc_state *cstate,
                             const struct drm_plane_state *pstate,
                             const int plane)
@@ -4151,6 +4151,7 @@ skl_plane_relative_data_rate(const struct intel_crtc_state *cstate,
        struct drm_framebuffer *fb;
        u32 format;
        uint_fixed_16_16_t down_scale_amount;
+       u64 rate;
 
        if (!intel_pstate->base.visible)
                return 0;
@@ -4177,28 +4178,26 @@ skl_plane_relative_data_rate(const struct intel_crtc_state *cstate,
                height /= 2;
        }
 
-       data_rate = width * height * fb->format->cpp[plane];
+       data_rate = width * height;
 
        down_scale_amount = skl_plane_downscale_amount(cstate, intel_pstate);
 
-       return mul_round_up_u32_fixed16(data_rate, down_scale_amount);
+       rate = mul_round_up_u32_fixed16(data_rate, down_scale_amount);
+
+       rate *= fb->format->cpp[plane];
+       return rate;
 }
 
-/*
- * We don't overflow 32 bits. Worst case is 3 planes enabled, each fetching
- * a 8192x4096@32bpp framebuffer:
- *   3 * 4096 * 8192  * 4 < 2^32
- */
-static unsigned int
+static u64
 skl_get_total_relative_data_rate(struct intel_crtc_state *intel_cstate,
-                                unsigned int *plane_data_rate,
-                                unsigned int *uv_plane_data_rate)
+                                u64 *plane_data_rate,
+                                u64 *uv_plane_data_rate)
 {
        struct drm_crtc_state *cstate = &intel_cstate->base;
        struct drm_atomic_state *state = cstate->state;
        struct drm_plane *plane;
        const struct drm_plane_state *pstate;
-       unsigned int total_data_rate = 0;
+       u64 total_data_rate = 0;
 
        if (WARN_ON(!state))
                return 0;
@@ -4206,7 +4205,7 @@ skl_get_total_relative_data_rate(struct intel_crtc_state *intel_cstate,
        /* Calculate and cache data rate for each plane */
        drm_atomic_crtc_state_for_each_plane_state(plane, pstate, cstate) {
                enum plane_id plane_id = to_intel_plane(plane)->id;
-               unsigned int rate;
+               u64 rate;
 
                /* packed/y */
                rate = skl_plane_relative_data_rate(intel_cstate,
@@ -4325,11 +4324,11 @@ skl_allocate_pipe_ddb(struct intel_crtc_state *cstate,
        uint16_t alloc_size, start;
        uint16_t minimum[I915_MAX_PLANES] = {};
        uint16_t uv_minimum[I915_MAX_PLANES] = {};
-       unsigned int total_data_rate;
+       u64 total_data_rate;
        enum plane_id plane_id;
        int num_active;
-       unsigned int plane_data_rate[I915_MAX_PLANES] = {};
-       unsigned int uv_plane_data_rate[I915_MAX_PLANES] = {};
+       u64 plane_data_rate[I915_MAX_PLANES] = {};
+       u64 uv_plane_data_rate[I915_MAX_PLANES] = {};
        uint16_t total_min_blocks = 0;
 
        /* Clear the partitioning for disabled planes. */
@@ -4388,7 +4387,7 @@ skl_allocate_pipe_ddb(struct intel_crtc_state *cstate,
 
        start = alloc->start;
        for_each_plane_id_on_crtc(intel_crtc, plane_id) {
-               unsigned int data_rate, uv_data_rate;
+               u64 data_rate, uv_data_rate;
                uint16_t plane_blocks, uv_plane_blocks;
 
                if (plane_id == PLANE_CURSOR)
@@ -4402,8 +4401,7 @@ skl_allocate_pipe_ddb(struct intel_crtc_state *cstate,
                 * result is < available as data_rate / total_data_rate < 1
                 */
                plane_blocks = minimum[plane_id];
-               plane_blocks += div_u64((uint64_t)alloc_size * data_rate,
-                                       total_data_rate);
+               plane_blocks += div64_u64(alloc_size * data_rate, total_data_rate);
 
                /* Leave disabled planes at (0,0) */
                if (data_rate) {
@@ -4417,8 +4415,7 @@ skl_allocate_pipe_ddb(struct intel_crtc_state *cstate,
                uv_data_rate = uv_plane_data_rate[plane_id];
 
                uv_plane_blocks = uv_minimum[plane_id];
-               uv_plane_blocks += div_u64((uint64_t)alloc_size * uv_data_rate,
-                                          total_data_rate);
+               uv_plane_blocks += div64_u64(alloc_size * uv_data_rate, total_data_rate);
 
                if (uv_data_rate) {
                        ddb->uv_plane[pipe][plane_id].start = start;