drm/i915: Replace hardcoded cacheline size with macro
authorChris Wilson <chris@chris-wilson.co.uk>
Wed, 9 Apr 2014 08:19:40 +0000 (09:19 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Mon, 5 May 2014 07:08:36 +0000 (09:08 +0200)
For readibility and guess at the meaning behind the constants.

v2: Claim only the meagerest connections with reality.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Oscar Mateo <oscar.mateo@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/intel_ringbuffer.c

index eb3dd26b94ded4b4c91e020108e8d0f9a6b5952d..890e0986edbe9ca8ab469c65b2fd67e169b91951 100644 (file)
 #include "i915_trace.h"
 #include "intel_drv.h"
 
+/* Early gen2 devices have a cacheline of just 32 bytes, using 64 is overkill,
+ * but keeps the logic simple. Indeed, the whole purpose of this macro is just
+ * to give some inclination as to some of the magic values used in the various
+ * workarounds!
+ */
+#define CACHELINE_BYTES 64
+
 static inline int ring_space(struct intel_ring_buffer *ring)
 {
        int space = (ring->head & HEAD_ADDR) - (ring->tail + I915_RING_FREE_SPACE);
@@ -179,7 +186,7 @@ gen4_render_ring_flush(struct intel_ring_buffer *ring,
 static int
 intel_emit_post_sync_nonzero_flush(struct intel_ring_buffer *ring)
 {
-       u32 scratch_addr = ring->scratch.gtt_offset + 128;
+       u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES;
        int ret;
 
 
@@ -216,7 +223,7 @@ gen6_render_ring_flush(struct intel_ring_buffer *ring,
                          u32 invalidate_domains, u32 flush_domains)
 {
        u32 flags = 0;
-       u32 scratch_addr = ring->scratch.gtt_offset + 128;
+       u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES;
        int ret;
 
        /* Force SNB workarounds for PIPE_CONTROL flushes */
@@ -310,7 +317,7 @@ gen7_render_ring_flush(struct intel_ring_buffer *ring,
                       u32 invalidate_domains, u32 flush_domains)
 {
        u32 flags = 0;
-       u32 scratch_addr = ring->scratch.gtt_offset + 128;
+       u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES;
        int ret;
 
        /*
@@ -371,7 +378,7 @@ gen8_render_ring_flush(struct intel_ring_buffer *ring,
                       u32 invalidate_domains, u32 flush_domains)
 {
        u32 flags = 0;
-       u32 scratch_addr = ring->scratch.gtt_offset + 128;
+       u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES;
        int ret;
 
        flags |= PIPE_CONTROL_CS_STALL;
@@ -783,7 +790,7 @@ do {                                                                        \
 static int
 pc_render_add_request(struct intel_ring_buffer *ring)
 {
-       u32 scratch_addr = ring->scratch.gtt_offset + 128;
+       u32 scratch_addr = ring->scratch.gtt_offset + 2 * CACHELINE_BYTES;
        int ret;
 
        /* For Ironlake, MI_USER_INTERRUPT was deprecated and apparently
@@ -805,15 +812,15 @@ pc_render_add_request(struct intel_ring_buffer *ring)
        intel_ring_emit(ring, ring->outstanding_lazy_seqno);
        intel_ring_emit(ring, 0);
        PIPE_CONTROL_FLUSH(ring, scratch_addr);
-       scratch_addr += 128; /* write to separate cachelines */
+       scratch_addr += 2 * CACHELINE_BYTES; /* write to separate cachelines */
        PIPE_CONTROL_FLUSH(ring, scratch_addr);
-       scratch_addr += 128;
+       scratch_addr += 2 * CACHELINE_BYTES;
        PIPE_CONTROL_FLUSH(ring, scratch_addr);
-       scratch_addr += 128;
+       scratch_addr += 2 * CACHELINE_BYTES;
        PIPE_CONTROL_FLUSH(ring, scratch_addr);
-       scratch_addr += 128;
+       scratch_addr += 2 * CACHELINE_BYTES;
        PIPE_CONTROL_FLUSH(ring, scratch_addr);
-       scratch_addr += 128;
+       scratch_addr += 2 * CACHELINE_BYTES;
        PIPE_CONTROL_FLUSH(ring, scratch_addr);
 
        intel_ring_emit(ring, GFX_OP_PIPE_CONTROL(4) | PIPE_CONTROL_QW_WRITE |
@@ -1422,7 +1429,7 @@ static int intel_init_ring_buffer(struct drm_device *dev,
         */
        ring->effective_size = ring->size;
        if (IS_I830(ring->dev) || IS_845G(ring->dev))
-               ring->effective_size -= 128;
+               ring->effective_size -= 2 * CACHELINE_BYTES;
 
        i915_cmd_parser_init_ring(ring);
 
@@ -1683,12 +1690,13 @@ int intel_ring_begin(struct intel_ring_buffer *ring,
 /* Align the ring tail to a cacheline boundary */
 int intel_ring_cacheline_align(struct intel_ring_buffer *ring)
 {
-       int num_dwords = (64 - (ring->tail & 63)) / sizeof(uint32_t);
+       int num_dwords = (ring->tail & (CACHELINE_BYTES - 1)) / sizeof(uint32_t);
        int ret;
 
        if (num_dwords == 0)
                return 0;
 
+       num_dwords = CACHELINE_BYTES / sizeof(uint32_t) - num_dwords;
        ret = intel_ring_begin(ring, num_dwords);
        if (ret)
                return ret;
@@ -2045,7 +2053,7 @@ int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size)
        ring->size = size;
        ring->effective_size = ring->size;
        if (IS_I830(ring->dev) || IS_845G(ring->dev))
-               ring->effective_size -= 128;
+               ring->effective_size -= 2 * CACHELINE_BYTES;
 
        ring->virtual_start = ioremap_wc(start, size);
        if (ring->virtual_start == NULL) {