Previously we only accommodated having a vma pinned by a small number of
users, with the maximum being pinned for use by the display engine. As
such, we used a small bitfield only large enough to allow the vma to
be pinned twice (for back/front buffers) in each scanout plane. Keeping
the maximum permissible pin_count small allows us to quickly catch a
potential leak. However, as we want to split a 4096B page into 64
different cachelines and pin each cacheline for use by a different
timeline, we will exceed the current maximum permissible vma->pin_count
and so time has come to enlarge it.
Whilst we are here, try to pull together the similar bits:
Address/layout specification:
- bias, mappable, zone_4g: address limit specifiers
- fixed: address override, limits still apply though
- high: not strictly an address limit, but an address direction to search
Search controls:
- nonblock, nonfault, noevict
v2: Rewrite the guideline comment on bit consumption.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Reviewed-by: John Harrison <john.C.Harrison@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_gem_gtt.h | 26 ++++++++---------
drivers/gpu/drm/i915/i915_vma.h | 45 +++++++++++++++++++----------
2 files changed, 42 insertions(+), 29 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.h b/drivers/gpu/drm/i915/i915_gem_gtt.h
index bd679c8c56dd..03ade71b8d9a 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.h
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.h
@@ -642,19 +642,19 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
/* Flags used by pin/bind&friends. */
#define PIN_NONBLOCK BIT_ULL(0)
-#define PIN_MAPPABLE BIT_ULL(1)
-#define PIN_ZONE_4G BIT_ULL(2)
-#define PIN_NONFAULT BIT_ULL(3)
-#define PIN_NOEVICT BIT_ULL(4)
-
-#define PIN_MBZ BIT_ULL(5) /* I915_VMA_PIN_OVERFLOW */
-#define PIN_GLOBAL BIT_ULL(6) /* I915_VMA_GLOBAL_BIND */
-#define PIN_USER BIT_ULL(7) /* I915_VMA_LOCAL_BIND */
-#define PIN_UPDATE BIT_ULL(8)
-
-#define PIN_HIGH BIT_ULL(9)
-#define PIN_OFFSET_BIAS BIT_ULL(10)
-#define PIN_OFFSET_FIXED BIT_ULL(11)
+#define PIN_NONFAULT BIT_ULL(1)
+#define PIN_NOEVICT BIT_ULL(2)
+#define PIN_MAPPABLE BIT_ULL(3)
+#define PIN_ZONE_4G BIT_ULL(4)
+#define PIN_HIGH BIT_ULL(5)
+#define PIN_OFFSET_BIAS BIT_ULL(6)
+#define PIN_OFFSET_FIXED BIT_ULL(7)
+
+#define PIN_MBZ BIT_ULL(8) /* I915_VMA_PIN_OVERFLOW */
+#define PIN_GLOBAL BIT_ULL(9) /* I915_VMA_GLOBAL_BIND */
+#define PIN_USER BIT_ULL(10) /* I915_VMA_LOCAL_BIND */
+#define PIN_UPDATE BIT_ULL(11)
+
#define PIN_OFFSET_MASK (-I915_GTT_PAGE_SIZE)
#endif
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 7252abc73d3e..5793abe509a2 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -71,29 +71,42 @@ struct i915_vma {
unsigned int open_count;
unsigned long flags;
/**
- * How many users have pinned this object in GTT space. The following
- * users can each hold at most one reference: pwrite/pread, execbuffer
- * (objects are not allowed multiple times for the same batchbuffer),
- * and the framebuffer code. When switching/pageflipping, the
- * framebuffer code has at most two buffers pinned per crtc.
+ * How many users have pinned this object in GTT space.
*
- * In the worst case this is 1 + 1 + 1 + 2*2 = 7. That would fit into 3
- * bits with absolutely no headroom. So use 4 bits.
+ * This is a tightly bound, fairly small number of users, so we
+ * stuff inside the flags field so that we can both check for overflow
+ * and detect a no-op i915_vma_pin() in a single check, while also
+ * pinning the vma.
+ *
+ * The worst case display setup would have the same vma pinned for
+ * use on each plane on each crtc, while also building the next atomic
+ * state and holding a pin for the length of the cleanup queue. In the
+ * future, the flip queue may be increased from 1.
+ * Estimated worst case: 3 [qlen] * 4 [max crtcs] * 7 [max planes] = 84
+ *
+ * For GEM, the number of concurrent users for pwrite/pread is
+ * unbounded. For execbuffer, it is currently one but will in future
+ * be extended to allow multiple clients to pin vma concurrently.
+ *
+ * We also use suballocated pages, with each suballocation claiming
+ * its own pin on the shared vma. At present, this is limited to
+ * exclusive cachelines of a single page, so a maximum of 64 possible
+ * users.
*/
-#define I915_VMA_PIN_MASK 0xf
-#define I915_VMA_PIN_OVERFLOW BIT(5)
+#define I915_VMA_PIN_MASK 0xff
+#define I915_VMA_PIN_OVERFLOW BIT(8)
/** Flags and address space this VMA is bound to */
-#define I915_VMA_GLOBAL_BIND BIT(6)
-#define I915_VMA_LOCAL_BIND BIT(7)
+#define I915_VMA_GLOBAL_BIND BIT(9)
+#define I915_VMA_LOCAL_BIND BIT(10)
#define I915_VMA_BIND_MASK (I915_VMA_GLOBAL_BIND | I915_VMA_LOCAL_BIND | I915_VMA_PIN_OVERFLOW)
-#define I915_VMA_GGTT BIT(8)
-#define I915_VMA_CAN_FENCE BIT(9)
-#define I915_VMA_CLOSED BIT(10)
-#define I915_VMA_USERFAULT_BIT 11
+#define I915_VMA_GGTT BIT(11)
+#define I915_VMA_CAN_FENCE BIT(12)
+#define I915_VMA_CLOSED BIT(13)
+#define I915_VMA_USERFAULT_BIT 14
#define I915_VMA_USERFAULT BIT(I915_VMA_USERFAULT_BIT)
-#define I915_VMA_GGTT_WRITE BIT(12)
+#define I915_VMA_GGTT_WRITE BIT(15)
unsigned int active_count;
struct rb_root active;
--
2.20.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
