drm/i915: Use two 32bit reads for select 64bit REG_READ ioctls

Since the hardware sometimes mysteriously totally flummoxes the 64bit
read of a 64bit register when read using a single instruction, split the
read into two instructions. Since the read here is of automatically
incrementing timestamp counters, we also have to be very careful in
order to make sure that it does not increment between the two
instructions.

However, since userspace tried to workaround this issue and so enshrined
this ABI for a broken hardware read and in the process neglected that
the read only fails in some environments, we have to introduce a new
uABI flag for userspace to request the 2x32 bit accurate read of the
timestamp.

v2: Fix alignment check and include details of the workaround for
userspace.

Reported-by: Karol Herbst <freedesktop@karolherbst.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91317
Testcase: igt/gem_reg_read
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: stable@vger.kernel.org
Tested-by: Michał Winiarski <michal.winiarski@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

1 files changed, 19 insertions, 7 deletions

diff --git a/drivers/gpu/drm/i915/intel_uncore.c b/drivers/gpu/drm/i915/intel_uncore.c
index a6d8a3ee7750..260389acfb77 100644
--- a/drivers/gpu/drm/i915/intel_uncore.c
+++ b/drivers/gpu/drm/i915/intel_uncore.c
@@ -1274,10 +1274,12 @@ int i915_reg_read_ioctl(struct drm_device *dev,
 	struct drm_i915_private *dev_priv = dev->dev_private;
 	struct drm_i915_reg_read *reg = data;
 	struct register_whitelist const *entry = whitelist;
+	unsigned size;
+	u64 offset;
 	int i, ret = 0;
 
 	for (i = 0; i < ARRAY_SIZE(whitelist); i++, entry++) {
-		if (entry->offset == reg->offset &&
+		if (entry->offset == (reg->offset & -entry->size) &&
 		    (1 << INTEL_INFO(dev)->gen & entry->gen_bitmask))
 			break;
 	}
@@ -1285,23 +1287,33 @@ int i915_reg_read_ioctl(struct drm_device *dev,
 	if (i == ARRAY_SIZE(whitelist))
 		return -EINVAL;
 
+	/* We use the low bits to encode extra flags as the register should
+	 * be naturally aligned (and those that are not so aligned merely
+	 * limit the available flags for that register).
+	 */
+	offset = entry->offset;
+	size = entry->size;
+	size |= reg->offset ^ offset;
+
 	intel_runtime_pm_get(dev_priv);
 
-	switch (entry->size) {
+	switch (size) {
+	case 8 | 1:
+		reg->val = I915_READ64_2x32(offset, offset+4);
+		break;
 	case 8:
-		reg->val = I915_READ64(reg->offset);
+		reg->val = I915_READ64(offset);
 		break;
 	case 4:
-		reg->val = I915_READ(reg->offset);
+		reg->val = I915_READ(offset);
 		break;
 	case 2:
-		reg->val = I915_READ16(reg->offset);
+		reg->val = I915_READ16(offset);
 		break;
 	case 1:
-		reg->val = I915_READ8(reg->offset);
+		reg->val = I915_READ8(offset);
 		break;
 	default:
-		MISSING_CASE(entry->size);
 		ret = -EINVAL;
 		goto out;
 	}