intel-gem: Add two new ioctls for managing tiling on objects.

Various chips have exciting interactions between the CPU and the GPU's
different ways of accessing interleaved memory, so we need some kernel
assistance in determining how it works.

Only fully tested on GM965 so far.

linux-core/Makefile.kernel

@@ -20,7 +20,8 @@ r128-objs   := r128_drv.o r128_cce.o r128_state.o r128_irq.o
 mga-objs    := mga_drv.o mga_dma.o mga_state.o mga_warp.o mga_irq.o
 i810-objs   := i810_drv.o i810_dma.o
 i915-objs   := i915_drv.o i915_dma.o i915_irq.o i915_mem.o i915_fence.o \
-		i915_buffer.o i915_compat.o i915_execbuf.o i915_gem.o
+		i915_buffer.o i915_compat.o i915_execbuf.o \
+		i915_gem.o i915_gem_tiling.o
 nouveau-objs := nouveau_drv.o nouveau_state.o nouveau_fifo.o nouveau_mem.o \
 		nouveau_object.o nouveau_irq.o nouveau_notifier.o nouveau_swmthd.o \
 		nouveau_sgdma.o nouveau_dma.o nouveau_bo.o nouveau_fence.o \

linux-core/i915_gem.c

@@ -2702,3 +2702,18 @@ i915_gem_lastclose(struct drm_device *dev)
 	
 	mutex_unlock(&dev->struct_mutex);
 }
+
+void i915_gem_load(struct drm_device *dev)
+{
+	drm_i915_private_t *dev_priv = dev->dev_private;
+
+	INIT_LIST_HEAD(&dev_priv->mm.active_list);
+	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
+	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
+	INIT_LIST_HEAD(&dev_priv->mm.request_list);
+	INIT_DELAYED_WORK(&dev_priv->mm.retire_work,
+			  i915_gem_retire_work_handler);
+	dev_priv->mm.next_gem_seqno = 1;
+
+	i915_gem_detect_bit_6_swizzle(dev);
+}

linux-core/i915_gem_tiling.c

@@ -0,0 +1,290 @@
+/*
+ * Copyright © 2008 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Eric Anholt <eric@anholt.net>
+ *
+ */
+
+#include "drmP.h"
+#include "drm.h"
+#include "i915_drm.h"
+#include "i915_drv.h"
+
+/** @file i915_gem_tiling.c
+ *
+ * Support for managing tiling state of buffer objects.
+ *
+ * The idea behind tiling is to increase cache hit rates by rearranging
+ * pixel data so that a group of pixel accesses are in the same cacheline.
+ * Performance improvement from doing this on the back/depth buffer are on
+ * the order of 30%.
+ *
+ * Intel architectures make this somewhat more complicated, though, by
+ * adjustments made to addressing of data when the memory is in interleaved
+ * mode (matched pairs of DIMMS) to improve memory bandwidth.
+ * For interleaved memory, the CPU sends every sequential 64 bytes
+ * to an alternate memory channel so it can get the bandwidth from both.
+ *
+ * The GPU also rearranges its accesses for increased bandwidth to interleaved
+ * memory, and it matches what the CPU does for non-tiled.  However, when tiled
+ * it does it a little differently, since one walks addresses not just in the
+ * X direction but also Y.  So, along with alternating channels when bit
+ * 6 of the address flips, it also alternates when other bits flip --  Bits 9
+ * (every 512 bytes, an X tile scanline) and 10 (every two X tile scanlines)
+ * are common to both the 915 and 965-class hardware.
+ *
+ * The CPU also sometimes XORs in higher bits as well, to improve
+ * bandwidth doing strided access like we do so frequently in graphics.  This
+ * is called "Channel XOR Randomization" in the MCH documentation.  The result
+ * is that the CPU is XORing in either bit 11 or bit 17 to bit 6 of its address
+ * decode.
+ *
+ * All of this bit 6 XORing has an effect on our memory management,
+ * as we need to make sure that the 3d driver can correctly address object
+ * contents.
+ *
+ * If we don't have interleaved memory, all tiling is safe and no swizzling is
+ * required.
+ *
+ * When bit 17 is XORed in, we simply refuse to tile at all.  Bit
+ * 17 is not just a page offset, so as we page an objet out and back in,
+ * individual pages in it will have different bit 17 addresses, resulting in
+ * each 64 bytes being swapped with its neighbor!
+ *
+ * Otherwise, if interleaved, we have to tell the 3d driver what the address
+ * swizzling it needs to do is, since it's writing with the CPU to the pages
+ * (bit 6 and potentially bit 11 XORed in), and the GPU is reading from the
+ * pages (bit 6, 9, and 10 XORed in), resulting in a cumulative bit swizzling
+ * required by the CPU of XORing in bit 6, 9, 10, and potentially 11, in order
+ * to match what the GPU expects.
+ */
+
+/**
+ * Detects bit 6 swizzling of address lookup between IGD access and CPU
+ * access through main memory.
+ */
+void
+i915_gem_detect_bit_6_swizzle(struct drm_device *dev)
+{
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct pci_dev *bridge;
+	uint32_t swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
+	uint32_t swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
+	int mchbar_offset;
+	volatile char *mchbar;
+	int ret;
+
+	bridge = pci_get_bus_and_slot(0, PCI_DEVFN(0,0));
+	if (bridge == NULL) {
+		DRM_ERROR("Couldn't get bridge device\n");
+		return;
+	}
+
+	ret = pci_enable_device(bridge);
+	if (ret != 0) {
+		DRM_ERROR("pci_enable_device failed: %d\n", ret);
+		return;
+	}
+
+	if (IS_I965G(dev))
+		mchbar_offset = 0x48;
+	else
+		mchbar_offset = 0x44;
+
+	/* Use resource 2 for our BAR that's stashed in a nonstandard location,
+	 * since the bridge would only ever use standard BARs 0-1 (though it
+	 * doesn't anyway)
+	 */
+	pci_read_base(bridge, mchbar_offset, &bridge->resource[2]);
+
+	mchbar = ioremap(pci_resource_start(bridge, 2),
+			 pci_resource_len(bridge, 2));
+	if (mchbar == NULL) {
+		DRM_ERROR("Couldn't map MCHBAR to determine tile swizzling\n");
+		return;
+	}
+
+	if (IS_I965G(dev) && !IS_I965GM(dev)) {
+		uint32_t chdecmisc;
+
+		/* On the 965, channel interleave appears to be determined by
+		 * the flex bit.  If flex is set, then the ranks (sides of a
+		 * DIMM) of memory will be "stacked" (physical addresses walk
+		 * through one rank then move on to the next, flipping channels
+		 * or not depending on rank configuration).  The GPU in this
+		 * case does exactly the same addressing as the CPU.
+		 *
+		 * Unlike the 945, channel randomization based does not
+		 * appear to be available.
+		 *
+		 * XXX: While the G965 doesn't appear to do any interleaving
+		 * when the DIMMs are not exactly matched, the G4x chipsets
+		 * might be for "L-shaped" configurations, and will need to be
+		 * detected.
+		 *
+		 * L-shaped configuration:
+		 *
+		 * +-----+
+		 * |     |
+		 * |DIMM2|         <-- non-interleaved
+		 * +-----+
+		 * +-----+ +-----+
+		 * |     | |     |
+		 * |DIMM0| |DIMM1| <-- interleaved area
+		 * +-----+ +-----+
+		 */
+		chdecmisc = readb(mchbar + CHDECMISC);
+
+		if (chdecmisc & CHDECMISC_FLEXMEMORY) {
+			swizzle_x = I915_BIT_6_SWIZZLE_NONE;
+			swizzle_y = I915_BIT_6_SWIZZLE_NONE;
+		} else {
+			swizzle_x = I915_BIT_6_SWIZZLE_9_10;
+			swizzle_y = I915_BIT_6_SWIZZLE_9;
+		}
+	} else if (IS_I9XX(dev)){
+		uint32_t dcc;
+
+		/* On 915-945 and GM965, channel interleave by the CPU is
+		 * determined by DCC.  The CPU will alternate based on bit 6
+		 * in interleaved mode, and the GPU will then also alternate
+		 * on bit 6, 9, and 10 for X, but the CPU may also optionally
+		 * alternate based on bit 17 (XOR not disabled and XOR
+		 * bit == 17).
+		 */
+		dcc = readl(mchbar + DCC);
+		switch (dcc & DCC_ADDRESSING_MODE_MASK) {
+		case DCC_ADDRESSING_MODE_SINGLE_CHANNEL:
+		case DCC_ADDRESSING_MODE_DUAL_CHANNEL_ASYMMETRIC:
+			swizzle_x = I915_BIT_6_SWIZZLE_NONE;
+			swizzle_y = I915_BIT_6_SWIZZLE_NONE;
+			break;
+		case DCC_ADDRESSING_MODE_DUAL_CHANNEL_INTERLEAVED:
+			if (IS_I915G(dev) || IS_I915GM(dev) ||
+			    dcc & DCC_CHANNEL_XOR_DISABLE) {
+				swizzle_x = I915_BIT_6_SWIZZLE_9_10;
+				swizzle_y = I915_BIT_6_SWIZZLE_9;
+			} else if (IS_I965GM(dev)) {
+				/* GM965 only does bit 11-based channel
+				 * randomization
+				 */
+				swizzle_x = I915_BIT_6_SWIZZLE_9_10_11;
+				swizzle_y = I915_BIT_6_SWIZZLE_9_11;
+			} else {
+				/* Bit 17 or perhaps other swizzling */
+				swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN;
+				swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN;
+			}
+			break;
+		}
+	}
+
+	iounmap(mchbar);
+
+	dev_priv->mm.bit_6_swizzle_x = swizzle_x;
+	dev_priv->mm.bit_6_swizzle_y = swizzle_y;
+}
+
+/**
+ * Sets the tiling mode of an object, returning the required swizzling of
+ * bit 6 of addresses in the object.
+ */
+int
+i915_gem_set_tiling(struct drm_device *dev, void *data,
+		   struct drm_file *file_priv)
+{
+	struct drm_i915_gem_set_tiling *args = data;
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct drm_gem_object *obj;
+	struct drm_i915_gem_object *obj_priv;
+
+	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
+	if (obj == NULL)
+		return -EINVAL;
+	obj_priv = obj->driver_private;
+
+	mutex_lock(&dev->struct_mutex);
+
+	if (args->tiling_mode == I915_TILING_NONE) {
+		obj_priv->tiling_mode = I915_TILING_NONE;
+		args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
+	} else {
+		if (args->tiling_mode == I915_TILING_X)
+			args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
+		else
+			args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
+		/* If we can't handle the swizzling, make it untiled. */
+		if (args->swizzle_mode == I915_BIT_6_SWIZZLE_UNKNOWN) {
+			args->tiling_mode = I915_TILING_NONE;
+			args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
+		}
+	}
+	obj_priv->tiling_mode = args->tiling_mode;
+
+	mutex_unlock(&dev->struct_mutex);
+
+	drm_gem_object_unreference(obj);
+
+	return 0;
+}
+
+/**
+ * Returns the current tiling mode and required bit 6 swizzling for the object.
+ */
+int
+i915_gem_get_tiling(struct drm_device *dev, void *data,
+		   struct drm_file *file_priv)
+{
+	struct drm_i915_gem_get_tiling *args = data;
+	drm_i915_private_t *dev_priv = dev->dev_private;
+	struct drm_gem_object *obj;
+	struct drm_i915_gem_object *obj_priv;
+
+	obj = drm_gem_object_lookup(dev, file_priv, args->handle);
+	if (obj == NULL)
+		return -EINVAL;
+	obj_priv = obj->driver_private;
+
+	mutex_lock(&dev->struct_mutex);
+
+	args->tiling_mode = obj_priv->tiling_mode;
+	switch (obj_priv->tiling_mode) {
+	case I915_TILING_X:
+		args->swizzle_mode = dev_priv->mm.bit_6_swizzle_x;
+		break;
+	case I915_TILING_Y:
+		args->swizzle_mode = dev_priv->mm.bit_6_swizzle_y;
+		break;
+	case I915_TILING_NONE:
+		args->swizzle_mode = I915_BIT_6_SWIZZLE_NONE;
+		break;
+	default:
+		DRM_ERROR("unknown tiling mode\n");
+	}
+
+	mutex_unlock(&dev->struct_mutex);
+
+	drm_gem_object_unreference(obj);
+
+	return 0;
+}

shared-core/i915_dma.c

@@ -1068,13 +1068,7 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
 	ret = drm_addmap(dev, base, size, _DRM_REGISTERS,
 		_DRM_KERNEL | _DRM_DRIVER, &dev_priv->mmio_map);
 
-	INIT_LIST_HEAD(&dev_priv->mm.active_list);
+	i915_gem_load(dev);
-	INIT_LIST_HEAD(&dev_priv->mm.flushing_list);
-	INIT_LIST_HEAD(&dev_priv->mm.inactive_list);
-	INIT_LIST_HEAD(&dev_priv->mm.request_list);
-	INIT_DELAYED_WORK(&dev_priv->mm.retire_work,
-			  i915_gem_retire_work_handler);
-	dev_priv->mm.next_gem_seqno = 1;
 
 #ifdef __linux__
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25)
@@ -1217,6 +1211,8 @@ struct drm_ioctl_desc i915_ioctls[] = {
 	DRM_IOCTL_DEF(DRM_I915_GEM_MMAP, i915_gem_mmap_ioctl, 0),
 	DRM_IOCTL_DEF(DRM_I915_GEM_SET_DOMAIN, i915_gem_set_domain_ioctl, 0),
 	DRM_IOCTL_DEF(DRM_I915_GEM_SW_FINISH, i915_gem_sw_finish_ioctl, 0),
+	DRM_IOCTL_DEF(DRM_I915_GEM_SET_TILING, i915_gem_set_tiling, 0),
+	DRM_IOCTL_DEF(DRM_I915_GEM_GET_TILING, i915_gem_get_tiling, 0),
 };
 
 int i915_max_ioctl = DRM_ARRAY_SIZE(i915_ioctls);

shared-core/i915_drm.h

@@ -190,6 +190,8 @@ typedef struct drm_i915_sarea {
 #define DRM_I915_GEM_MMAP	0x1e
 #define DRM_I915_GEM_SET_DOMAIN	0x1f
 #define DRM_I915_GEM_SW_FINISH	0x20
+#define DRM_I915_GEM_SET_TILING	0x21
+#define DRM_I915_GEM_GET_TILING	0x22
 
 #define DRM_IOCTL_I915_INIT		DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t)
 #define DRM_IOCTL_I915_FLUSH		DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH)
@@ -223,6 +225,8 @@ typedef struct drm_i915_sarea {
 #define DRM_IOCTL_I915_GEM_MMAP		DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP, struct drm_i915_gem_mmap)
 #define DRM_IOCTL_I915_GEM_SET_DOMAIN	DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SET_DOMAIN, struct drm_i915_gem_set_domain)
 #define DRM_IOCTL_I915_GEM_SW_FINISH	DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SW_FINISH, struct drm_i915_gem_sw_finish)
+#define DRM_IOCTL_I915_GEM_SET_TILING	DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_SET_TILING, struct drm_i915_gem_set_tiling)
+#define DRM_IOCTL_I915_GEM_GET_TILING	DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_GET_TILING, struct drm_i915_gem_get_tiling)
 
 /* Asynchronous page flipping:
  */
@@ -651,4 +655,64 @@ struct drm_i915_gem_busy {
 	uint32_t busy;
 };
 
+#define I915_TILING_NONE	0
+#define I915_TILING_X		1
+#define I915_TILING_Y		2
+
+#define I915_BIT_6_SWIZZLE_NONE		0
+#define I915_BIT_6_SWIZZLE_9		1
+#define I915_BIT_6_SWIZZLE_9_10		2
+#define I915_BIT_6_SWIZZLE_9_11		3
+#define I915_BIT_6_SWIZZLE_9_10_11	4
+/* Not seen by userland */
+#define I915_BIT_6_SWIZZLE_UNKNOWN	5
+
+struct drm_i915_gem_set_tiling {
+	/** Handle of the buffer to have its tiling state updated */
+	uint32_t handle;
+
+	/**
+	 * Tiling mode for the object (I915_TILING_NONE, I915_TILING_X,
+	 * I915_TILING_Y).
+	 *
+	 * This value is to be set on request, and will be updated by the
+	 * kernel on successful return with the actual chosen tiling layout.
+	 *
+	 * The tiling mode may be demoted to I915_TILING_NONE when the system
+	 * has bit 6 swizzling that can't be managed correctly by GEM.
+	 *
+	 * Buffer contents become undefined when changing tiling_mode.
+	 */
+	uint32_t tiling_mode;
+
+	/**
+	 * Stride in bytes for the object when in I915_TILING_X or
+	 * I915_TILING_Y.
+	 */
+	uint32_t stride;
+
+	/**
+	 * Returned address bit 6 swizzling required for CPU access through
+	 * mmap mapping.
+	 */
+	uint32_t swizzle_mode;
+};
+
+struct drm_i915_gem_get_tiling {
+	/** Handle of the buffer to get tiling state for. */
+	uint32_t handle;
+
+	/**
+	 * Current tiling mode for the object (I915_TILING_NONE, I915_TILING_X,
+	 * I915_TILING_Y).
+	 */
+	uint32_t tiling_mode;
+
+	/**
+	 * Returned address bit 6 swizzling required for CPU access through
+	 * mmap mapping.
+	 */
+	uint32_t swizzle_mode;
+};
+
 #endif				/* _I915_DRM_H_ */

shared-core/i915_drv.h

@@ -316,6 +316,11 @@ typedef struct drm_i915_private {
 		 * every pending request fail
 		 */
 		int wedged;
+
+		/** Bit 6 swizzling required for X tiling */
+		uint32_t bit_6_swizzle_x;
+		/** Bit 6 swizzling required for Y tiling */
+		uint32_t bit_6_swizzle_y;
 	} mm;
 } drm_i915_private_t;
 
@@ -376,6 +381,9 @@ struct drm_i915_gem_object {
 
 	/** Breadcrumb of last rendering to the buffer. */
 	uint32_t last_rendering_seqno;
+
+	/** Current tiling mode for the object. */
+	uint32_t tiling_mode;
 };
 
 /**
@@ -517,6 +525,11 @@ int i915_gem_entervt_ioctl(struct drm_device *dev, void *data,
 			   struct drm_file *file_priv);
 int i915_gem_leavevt_ioctl(struct drm_device *dev, void *data,
 			   struct drm_file *file_priv);
+int i915_gem_set_tiling(struct drm_device *dev, void *data,
+			struct drm_file *file_priv);
+int i915_gem_get_tiling(struct drm_device *dev, void *data,
+			struct drm_file *file_priv);
+void i915_gem_load(struct drm_device *dev);
 int i915_gem_proc_init(struct drm_minor *minor);
 void i915_gem_proc_cleanup(struct drm_minor *minor);
 int i915_gem_init_object(struct drm_gem_object *obj);
@@ -529,6 +542,9 @@ void i915_gem_retire_requests(struct drm_device *dev);
 void i915_gem_retire_work_handler(struct work_struct *work);
 #endif
 
+/* i915_gem_tiling.c */
+void i915_gem_detect_bit_6_swizzle(struct drm_device *dev);
+
 #ifdef __linux__
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25)
 extern void intel_init_chipset_flush_compat(struct drm_device *dev);
@@ -585,6 +601,19 @@ void i915_ring_validate(struct drm_device *dev, const char *func, int line);
 
 extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller);
 
+/* MCH MMIO space */
+/** 915-945 and GM965 MCH register controlling DRAM channel access */
+#define DCC		0x200
+#define DCC_ADDRESSING_MODE_SINGLE_CHANNEL		(0 << 0)
+#define DCC_ADDRESSING_MODE_DUAL_CHANNEL_ASYMMETRIC	(1 << 0)
+#define DCC_ADDRESSING_MODE_DUAL_CHANNEL_INTERLEAVED	(2 << 0)
+#define DCC_ADDRESSING_MODE_MASK			(3 << 0)
+#define DCC_CHANNEL_XOR_DISABLE				(1 << 10)
+
+/** 965 MCH register controlling DRAM channel configuration */
+#define CHDECMISC		0x111
+#define CHDECMISC_FLEXMEMORY		(1 << 1)
+
 /* Extended config space */
 #define LBB 0xf4