diff options
author | Eric Anholt <eric@anholt.net> | 2008-07-30 12:06:12 -0700 |
---|---|---|
committer | Dave Airlie <airlied@linux.ie> | 2008-10-18 07:10:12 +1000 |
commit | 673a394b1e3b69be886ff24abfd6df97c52e8d08 (patch) | |
tree | 61ca8299333ab50ffc46cf328b20eb25133392ff /drivers/gpu/drm | |
parent | d1d8c925b71dd6753bf438f9e14a9e5c5183bcc6 (diff) | |
download | linux-stable-673a394b1e3b69be886ff24abfd6df97c52e8d08.tar.gz linux-stable-673a394b1e3b69be886ff24abfd6df97c52e8d08.tar.bz2 linux-stable-673a394b1e3b69be886ff24abfd6df97c52e8d08.zip |
drm: Add GEM ("graphics execution manager") to i915 driver.
GEM allows the creation of persistent buffer objects accessible by the
graphics device through new ioctls for managing execution of commands on the
device. The userland API is almost entirely driver-specific to ensure that
any driver building on this model can easily map the interface to individual
driver requirements.
GEM is used by the 2d driver for managing its internal state allocations and
will be used for pixmap storage to reduce memory consumption and enable
zero-copy GLX_EXT_texture_from_pixmap, and in the 3d driver is used to enable
GL_EXT_framebuffer_object and GL_ARB_pixel_buffer_object.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Diffstat (limited to 'drivers/gpu/drm')
-rw-r--r-- | drivers/gpu/drm/Makefile | 5 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_agpsupport.c | 51 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_cache.c | 76 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_drv.c | 4 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_fops.c | 6 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_gem.c | 420 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_memory.c | 2 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_mm.c | 5 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_proc.c | 135 | ||||
-rw-r--r-- | drivers/gpu/drm/drm_stub.c | 10 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/Makefile | 6 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_dma.c | 94 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_drv.c | 11 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_drv.h | 253 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_gem.c | 2497 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_gem_debug.c | 201 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_gem_proc.c | 292 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_gem_tiling.c | 256 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_irq.c | 19 | ||||
-rw-r--r-- | drivers/gpu/drm/i915/i915_reg.h | 37 |
20 files changed, 4318 insertions, 62 deletions
diff --git a/drivers/gpu/drm/Makefile b/drivers/gpu/drm/Makefile index e9f9a97ae00a..74da99495e21 100644 --- a/drivers/gpu/drm/Makefile +++ b/drivers/gpu/drm/Makefile @@ -4,8 +4,9 @@ ccflags-y := -Iinclude/drm -drm-y := drm_auth.o drm_bufs.o drm_context.o drm_dma.o drm_drawable.o \ - drm_drv.o drm_fops.o drm_ioctl.o drm_irq.o \ +drm-y := drm_auth.o drm_bufs.o drm_cache.o \ + drm_context.o drm_dma.o drm_drawable.o \ + drm_drv.o drm_fops.o drm_gem.o drm_ioctl.o drm_irq.o \ drm_lock.o drm_memory.o drm_proc.o drm_stub.o drm_vm.o \ drm_agpsupport.o drm_scatter.o ati_pcigart.o drm_pci.o \ drm_sysfs.o drm_hashtab.o drm_sman.o drm_mm.o diff --git a/drivers/gpu/drm/drm_agpsupport.c b/drivers/gpu/drm/drm_agpsupport.c index aefa5ac4c0b1..2639be2db9e5 100644 --- a/drivers/gpu/drm/drm_agpsupport.c +++ b/drivers/gpu/drm/drm_agpsupport.c @@ -33,6 +33,7 @@ #include "drmP.h" #include <linux/module.h> +#include <asm/agp.h> #if __OS_HAS_AGP @@ -452,4 +453,52 @@ int drm_agp_unbind_memory(DRM_AGP_MEM * handle) return agp_unbind_memory(handle); } -#endif /* __OS_HAS_AGP */ +/** + * Binds a collection of pages into AGP memory at the given offset, returning + * the AGP memory structure containing them. + * + * No reference is held on the pages during this time -- it is up to the + * caller to handle that. + */ +DRM_AGP_MEM * +drm_agp_bind_pages(struct drm_device *dev, + struct page **pages, + unsigned long num_pages, + uint32_t gtt_offset) +{ + DRM_AGP_MEM *mem; + int ret, i; + + DRM_DEBUG("\n"); + + mem = drm_agp_allocate_memory(dev->agp->bridge, num_pages, + AGP_USER_MEMORY); + if (mem == NULL) { + DRM_ERROR("Failed to allocate memory for %ld pages\n", + num_pages); + return NULL; + } + + for (i = 0; i < num_pages; i++) + mem->memory[i] = phys_to_gart(page_to_phys(pages[i])); + mem->page_count = num_pages; + + mem->is_flushed = true; + ret = drm_agp_bind_memory(mem, gtt_offset / PAGE_SIZE); + if (ret != 0) { + DRM_ERROR("Failed to bind AGP memory: %d\n", ret); + agp_free_memory(mem); + return NULL; + } + + return mem; +} +EXPORT_SYMBOL(drm_agp_bind_pages); + +void drm_agp_chipset_flush(struct drm_device *dev) +{ + agp_flush_chipset(dev->agp->bridge); +} +EXPORT_SYMBOL(drm_agp_chipset_flush); + +#endif /* __OS_HAS_AGP */ diff --git a/drivers/gpu/drm/drm_cache.c b/drivers/gpu/drm/drm_cache.c new file mode 100644 index 000000000000..9475f7d9901d --- /dev/null +++ b/drivers/gpu/drm/drm_cache.c @@ -0,0 +1,76 @@ +/************************************************************************** + * + * Copyright (c) 2006-2007 Tungsten Graphics, Inc., Cedar Park, TX., USA + * All Rights Reserved. + * + * 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, sub license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL + * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS 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: Thomas Hellström <thomas-at-tungstengraphics-dot-com> + */ + +#include "drmP.h" + +#if defined(CONFIG_X86) +static void +drm_clflush_page(struct page *page) +{ + uint8_t *page_virtual; + unsigned int i; + + if (unlikely(page == NULL)) + return; + + page_virtual = kmap_atomic(page, KM_USER0); + for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size) + clflush(page_virtual + i); + kunmap_atomic(page_virtual, KM_USER0); +} +#endif + +static void +drm_clflush_ipi_handler(void *null) +{ + wbinvd(); +} + +void +drm_clflush_pages(struct page *pages[], unsigned long num_pages) +{ + +#if defined(CONFIG_X86) + if (cpu_has_clflush) { + unsigned long i; + + mb(); + for (i = 0; i < num_pages; ++i) + drm_clflush_page(*pages++); + mb(); + + return; + } +#endif + + if (on_each_cpu(drm_clflush_ipi_handler, NULL, 1) != 0) + DRM_ERROR("Timed out waiting for cache flush.\n"); +} +EXPORT_SYMBOL(drm_clflush_pages); diff --git a/drivers/gpu/drm/drm_drv.c b/drivers/gpu/drm/drm_drv.c index fb45fe7aeb8c..96f416afc3f6 100644 --- a/drivers/gpu/drm/drm_drv.c +++ b/drivers/gpu/drm/drm_drv.c @@ -119,6 +119,10 @@ static struct drm_ioctl_desc drm_ioctls[] = { DRM_IOCTL_DEF(DRM_IOCTL_MODESET_CTL, drm_modeset_ctl, 0), DRM_IOCTL_DEF(DRM_IOCTL_UPDATE_DRAW, drm_update_drawable_info, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY), + + DRM_IOCTL_DEF(DRM_IOCTL_GEM_CLOSE, drm_gem_close_ioctl, 0), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_FLINK, drm_gem_flink_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_OPEN, drm_gem_open_ioctl, DRM_AUTH), }; #define DRM_CORE_IOCTL_COUNT ARRAY_SIZE( drm_ioctls ) diff --git a/drivers/gpu/drm/drm_fops.c b/drivers/gpu/drm/drm_fops.c index dcf8b4dc9549..0d46627663b1 100644 --- a/drivers/gpu/drm/drm_fops.c +++ b/drivers/gpu/drm/drm_fops.c @@ -256,6 +256,9 @@ static int drm_open_helper(struct inode *inode, struct file *filp, INIT_LIST_HEAD(&priv->lhead); + if (dev->driver->driver_features & DRIVER_GEM) + drm_gem_open(dev, priv); + if (dev->driver->open) { ret = dev->driver->open(dev, priv); if (ret < 0) @@ -400,6 +403,9 @@ int drm_release(struct inode *inode, struct file *filp) dev->driver->reclaim_buffers(dev, file_priv); } + if (dev->driver->driver_features & DRIVER_GEM) + drm_gem_release(dev, file_priv); + drm_fasync(-1, filp, 0); mutex_lock(&dev->ctxlist_mutex); diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c new file mode 100644 index 000000000000..434155b387e9 --- /dev/null +++ b/drivers/gpu/drm/drm_gem.c @@ -0,0 +1,420 @@ +/* + * 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 <linux/types.h> +#include <linux/slab.h> +#include <linux/mm.h> +#include <linux/uaccess.h> +#include <linux/fs.h> +#include <linux/file.h> +#include <linux/module.h> +#include <linux/mman.h> +#include <linux/pagemap.h> +#include "drmP.h" + +/** @file drm_gem.c + * + * This file provides some of the base ioctls and library routines for + * the graphics memory manager implemented by each device driver. + * + * Because various devices have different requirements in terms of + * synchronization and migration strategies, implementing that is left up to + * the driver, and all that the general API provides should be generic -- + * allocating objects, reading/writing data with the cpu, freeing objects. + * Even there, platform-dependent optimizations for reading/writing data with + * the CPU mean we'll likely hook those out to driver-specific calls. However, + * the DRI2 implementation wants to have at least allocate/mmap be generic. + * + * The goal was to have swap-backed object allocation managed through + * struct file. However, file descriptors as handles to a struct file have + * two major failings: + * - Process limits prevent more than 1024 or so being used at a time by + * default. + * - Inability to allocate high fds will aggravate the X Server's select() + * handling, and likely that of many GL client applications as well. + * + * This led to a plan of using our own integer IDs (called handles, following + * DRM terminology) to mimic fds, and implement the fd syscalls we need as + * ioctls. The objects themselves will still include the struct file so + * that we can transition to fds if the required kernel infrastructure shows + * up at a later date, and as our interface with shmfs for memory allocation. + */ + +/** + * Initialize the GEM device fields + */ + +int +drm_gem_init(struct drm_device *dev) +{ + spin_lock_init(&dev->object_name_lock); + idr_init(&dev->object_name_idr); + atomic_set(&dev->object_count, 0); + atomic_set(&dev->object_memory, 0); + atomic_set(&dev->pin_count, 0); + atomic_set(&dev->pin_memory, 0); + atomic_set(&dev->gtt_count, 0); + atomic_set(&dev->gtt_memory, 0); + return 0; +} + +/** + * Allocate a GEM object of the specified size with shmfs backing store + */ +struct drm_gem_object * +drm_gem_object_alloc(struct drm_device *dev, size_t size) +{ + struct drm_gem_object *obj; + + BUG_ON((size & (PAGE_SIZE - 1)) != 0); + + obj = kcalloc(1, sizeof(*obj), GFP_KERNEL); + + obj->dev = dev; + obj->filp = shmem_file_setup("drm mm object", size, 0); + if (IS_ERR(obj->filp)) { + kfree(obj); + return NULL; + } + + kref_init(&obj->refcount); + kref_init(&obj->handlecount); + obj->size = size; + if (dev->driver->gem_init_object != NULL && + dev->driver->gem_init_object(obj) != 0) { + fput(obj->filp); + kfree(obj); + return NULL; + } + atomic_inc(&dev->object_count); + atomic_add(obj->size, &dev->object_memory); + return obj; +} +EXPORT_SYMBOL(drm_gem_object_alloc); + +/** + * Removes the mapping from handle to filp for this object. + */ +static int +drm_gem_handle_delete(struct drm_file *filp, int handle) +{ + struct drm_device *dev; + struct drm_gem_object *obj; + + /* This is gross. The idr system doesn't let us try a delete and + * return an error code. It just spews if you fail at deleting. + * So, we have to grab a lock around finding the object and then + * doing the delete on it and dropping the refcount, or the user + * could race us to double-decrement the refcount and cause a + * use-after-free later. Given the frequency of our handle lookups, + * we may want to use ida for number allocation and a hash table + * for the pointers, anyway. + */ + spin_lock(&filp->table_lock); + + /* Check if we currently have a reference on the object */ + obj = idr_find(&filp->object_idr, handle); + if (obj == NULL) { + spin_unlock(&filp->table_lock); + return -EINVAL; + } + dev = obj->dev; + + /* Release reference and decrement refcount. */ + idr_remove(&filp->object_idr, handle); + spin_unlock(&filp->table_lock); + + mutex_lock(&dev->struct_mutex); + drm_gem_object_handle_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +/** + * Create a handle for this object. This adds a handle reference + * to the object, which includes a regular reference count. Callers + * will likely want to dereference the object afterwards. + */ +int +drm_gem_handle_create(struct drm_file *file_priv, + struct drm_gem_object *obj, + int *handlep) +{ + int ret; + + /* + * Get the user-visible handle using idr. + */ +again: + /* ensure there is space available to allocate a handle */ + if (idr_pre_get(&file_priv->object_idr, GFP_KERNEL) == 0) + return -ENOMEM; + + /* do the allocation under our spinlock */ + spin_lock(&file_priv->table_lock); + ret = idr_get_new_above(&file_priv->object_idr, obj, 1, handlep); + spin_unlock(&file_priv->table_lock); + if (ret == -EAGAIN) + goto again; + + if (ret != 0) + return ret; + + drm_gem_object_handle_reference(obj); + return 0; +} +EXPORT_SYMBOL(drm_gem_handle_create); + +/** Returns a reference to the object named by the handle. */ +struct drm_gem_object * +drm_gem_object_lookup(struct drm_device *dev, struct drm_file *filp, + int handle) +{ + struct drm_gem_object *obj; + + spin_lock(&filp->table_lock); + + /* Check if we currently have a reference on the object */ + obj = idr_find(&filp->object_idr, handle); + if (obj == NULL) { + spin_unlock(&filp->table_lock); + return NULL; + } + + drm_gem_object_reference(obj); + + spin_unlock(&filp->table_lock); + + return obj; +} +EXPORT_SYMBOL(drm_gem_object_lookup); + +/** + * Releases the handle to an mm object. + */ +int +drm_gem_close_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_close *args = data; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + ret = drm_gem_handle_delete(file_priv, args->handle); + + return ret; +} + +/** + * Create a global name for an object, returning the name. + * + * Note that the name does not hold a reference; when the object + * is freed, the name goes away. + */ +int +drm_gem_flink_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_flink *args = data; + struct drm_gem_object *obj; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + +again: + if (idr_pre_get(&dev->object_name_idr, GFP_KERNEL) == 0) + return -ENOMEM; + + spin_lock(&dev->object_name_lock); + if (obj->name) { + spin_unlock(&dev->object_name_lock); + return -EEXIST; + } + ret = idr_get_new_above(&dev->object_name_idr, obj, 1, + &obj->name); + spin_unlock(&dev->object_name_lock); + if (ret == -EAGAIN) + goto again; + + if (ret != 0) { + mutex_lock(&dev->struct_mutex); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + + /* + * Leave the reference from the lookup around as the + * name table now holds one + */ + args->name = (uint64_t) obj->name; + + return 0; +} + +/** + * Open an object using the global name, returning a handle and the size. + * + * This handle (of course) holds a reference to the object, so the object + * will not go away until the handle is deleted. + */ +int +drm_gem_open_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_open *args = data; + struct drm_gem_object *obj; + int ret; + int handle; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + spin_lock(&dev->object_name_lock); + obj = idr_find(&dev->object_name_idr, (int) args->name); + if (obj) + drm_gem_object_reference(obj); + spin_unlock(&dev->object_name_lock); + if (!obj) + return -ENOENT; + + ret = drm_gem_handle_create(file_priv, obj, &handle); + mutex_lock(&dev->struct_mutex); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (ret) + return ret; + + args->handle = handle; + args->size = obj->size; + + return 0; +} + +/** + * Called at device open time, sets up the structure for handling refcounting + * of mm objects. + */ +void +drm_gem_open(struct drm_device *dev, struct drm_file *file_private) +{ + idr_init(&file_private->object_idr); + spin_lock_init(&file_private->table_lock); +} + +/** + * Called at device close to release the file's + * handle references on objects. + */ +static int +drm_gem_object_release_handle(int id, void *ptr, void *data) +{ + struct drm_gem_object *obj = ptr; + + drm_gem_object_handle_unreference(obj); + + return 0; +} + +/** + * Called at close time when the filp is going away. + * + * Releases any remaining references on objects by this filp. + */ +void +drm_gem_release(struct drm_device *dev, struct drm_file *file_private) +{ + mutex_lock(&dev->struct_mutex); + idr_for_each(&file_private->object_idr, + &drm_gem_object_release_handle, NULL); + + idr_destroy(&file_private->object_idr); + mutex_unlock(&dev->struct_mutex); +} + +/** + * Called after the last reference to the object has been lost. + * + * Frees the object + */ +void +drm_gem_object_free(struct kref *kref) +{ + struct drm_gem_object *obj = (struct drm_gem_object *) kref; + struct drm_device *dev = obj->dev; + + BUG_ON(!mutex_is_locked(&dev->struct_mutex)); + + if (dev->driver->gem_free_object != NULL) + dev->driver->gem_free_object(obj); + + fput(obj->filp); + atomic_dec(&dev->object_count); + atomic_sub(obj->size, &dev->object_memory); + kfree(obj); +} +EXPORT_SYMBOL(drm_gem_object_free); + +/** + * Called after the last handle to the object has been closed + * + * Removes any name for the object. Note that this must be + * called before drm_gem_object_free or we'll be touching + * freed memory + */ +void +drm_gem_object_handle_free(struct kref *kref) +{ + struct drm_gem_object *obj = container_of(kref, + struct drm_gem_object, + handlecount); + struct drm_device *dev = obj->dev; + + /* Remove any name for this object */ + spin_lock(&dev->object_name_lock); + if (obj->name) { + idr_remove(&dev->object_name_idr, obj->name); + spin_unlock(&dev->object_name_lock); + /* + * The object name held a reference to this object, drop + * that now. + */ + drm_gem_object_unreference(obj); + } else + spin_unlock(&dev->object_name_lock); + +} +EXPORT_SYMBOL(drm_gem_object_handle_free); + diff --git a/drivers/gpu/drm/drm_memory.c b/drivers/gpu/drm/drm_memory.c index 0177012845c6..803bc9e7ce3c 100644 --- a/drivers/gpu/drm/drm_memory.c +++ b/drivers/gpu/drm/drm_memory.c @@ -133,6 +133,7 @@ int drm_free_agp(DRM_AGP_MEM * handle, int pages) { return drm_agp_free_memory(handle) ? 0 : -EINVAL; } +EXPORT_SYMBOL(drm_free_agp); /** Wrapper around agp_bind_memory() */ int drm_bind_agp(DRM_AGP_MEM * handle, unsigned int start) @@ -145,6 +146,7 @@ int drm_unbind_agp(DRM_AGP_MEM * handle) { return drm_agp_unbind_memory(handle); } +EXPORT_SYMBOL(drm_unbind_agp); #else /* __OS_HAS_AGP */ static inline void *agp_remap(unsigned long offset, unsigned long size, diff --git a/drivers/gpu/drm/drm_mm.c b/drivers/gpu/drm/drm_mm.c index dcff9e9b52e3..217ad7dc7076 100644 --- a/drivers/gpu/drm/drm_mm.c +++ b/drivers/gpu/drm/drm_mm.c @@ -169,6 +169,7 @@ struct drm_mm_node *drm_mm_get_block(struct drm_mm_node * parent, return child; } +EXPORT_SYMBOL(drm_mm_get_block); /* * Put a block. Merge with the previous and / or next block if they are free. @@ -217,6 +218,7 @@ void drm_mm_put_block(struct drm_mm_node * cur) drm_free(cur, sizeof(*cur), DRM_MEM_MM); } } +EXPORT_SYMBOL(drm_mm_put_block); struct drm_mm_node *drm_mm_search_free(const struct drm_mm * mm, unsigned long size, @@ -265,6 +267,7 @@ int drm_mm_clean(struct drm_mm * mm) return (head->next->next == head); } +EXPORT_SYMBOL(drm_mm_search_free); int drm_mm_init(struct drm_mm * mm, unsigned long start, unsigned long size) { @@ -273,7 +276,7 @@ int drm_mm_init(struct drm_mm * mm, unsigned long start, unsigned long size) return drm_mm_create_tail_node(mm, start, size); } - +EXPORT_SYMBOL(drm_mm_init); void drm_mm_takedown(struct drm_mm * mm) { diff --git a/drivers/gpu/drm/drm_proc.c b/drivers/gpu/drm/drm_proc.c index 93b1e0475c93..d490db4c0de0 100644 --- a/drivers/gpu/drm/drm_proc.c +++ b/drivers/gpu/drm/drm_proc.c @@ -49,6 +49,10 @@ static int drm_queues_info(char *buf, char **start, off_t offset, int request, int *eof, void *data); static int drm_bufs_info(char *buf, char **start, off_t offset, int request, int *eof, void *data); +static int drm_gem_name_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data); +static int drm_gem_object_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data); #if DRM_DEBUG_CODE static int drm_vma_info(char *buf, char **start, off_t offset, int request, int *eof, void *data); @@ -60,13 +64,16 @@ static int drm_vma_info(char *buf, char **start, off_t offset, static struct drm_proc_list { const char *name; /**< file name */ int (*f) (char *, char **, off_t, int, int *, void *); /**< proc callback*/ + u32 driver_features; /**< Required driver features for this entry */ } drm_proc_list[] = { - {"name", drm_name_info}, - {"mem", drm_mem_info}, - {"vm", drm_vm_info}, - {"clients", drm_clients_info}, - {"queues", drm_queues_info}, - {"bufs", drm_bufs_info}, + {"name", drm_name_info, 0}, + {"mem", drm_mem_info, 0}, + {"vm", drm_vm_info, 0}, + {"clients", drm_clients_info, 0}, + {"queues", drm_queues_info, 0}, + {"bufs", drm_bufs_info, 0}, + {"gem_names", drm_gem_name_info, DRIVER_GEM}, + {"gem_objects", drm_gem_object_info, DRIVER_GEM}, #if DRM_DEBUG_CODE {"vma", drm_vma_info}, #endif @@ -90,8 +97,9 @@ static struct drm_proc_list { int drm_proc_init(struct drm_minor *minor, int minor_id, struct proc_dir_entry *root) { + struct drm_device *dev = minor->dev; struct proc_dir_entry *ent; - int i, j; + int i, j, ret; char name[64]; sprintf(name, "%d", minor_id); @@ -102,23 +110,42 @@ int drm_proc_init(struct drm_minor *minor, int minor_id, } for (i = 0; i < DRM_PROC_ENTRIES; i++) { + u32 features = drm_proc_list[i].driver_features; + + if (features != 0 && + (dev->driver->driver_features & features) != features) + continue; + ent = create_proc_entry(drm_proc_list[i].name, S_IFREG | S_IRUGO, minor->dev_root); if (!ent) { DRM_ERROR("Cannot create /proc/dri/%s/%s\n", name, drm_proc_list[i].name); - for (j = 0; j < i; j++) - remove_proc_entry(drm_proc_list[i].name, - minor->dev_root); - remove_proc_entry(name, root); - minor->dev_root = NULL; - return -1; + ret = -1; + goto fail; } ent->read_proc = drm_proc_list[i].f; ent->data = minor; } + if (dev->driver->proc_init) { + ret = dev->driver->proc_init(minor); + if (ret) { + DRM_ERROR("DRM: Driver failed to initialize " + "/proc/dri.\n"); + goto fail; + } + } + return 0; + fail: + + for (j = 0; j < i; j++) + remove_proc_entry(drm_proc_list[i].name, + minor->dev_root); + remove_proc_entry(name, root); + minor->dev_root = NULL; + return ret; } /** @@ -133,12 +160,16 @@ int drm_proc_init(struct drm_minor *minor, int minor_id, */ int drm_proc_cleanup(struct drm_minor *minor, struct proc_dir_entry *root) { + struct drm_device *dev = minor->dev; int i; char name[64]; if (!root || !minor->dev_root) return 0; + if (dev->driver->proc_cleanup) + dev->driver->proc_cleanup(minor); + for (i = 0; i < DRM_PROC_ENTRIES; i++) remove_proc_entry(drm_proc_list[i].name, minor->dev_root); sprintf(name, "%d", minor->index); @@ -480,6 +511,84 @@ static int drm_clients_info(char *buf, char **start, off_t offset, return ret; } +struct drm_gem_name_info_data { + int len; + char *buf; + int eof; +}; + +static int drm_gem_one_name_info(int id, void *ptr, void *data) +{ + struct drm_gem_object *obj = ptr; + struct drm_gem_name_info_data *nid = data; + + DRM_INFO("name %d size %d\n", obj->name, obj->size); + if (nid->eof) + return 0; + + nid->len += sprintf(&nid->buf[nid->len], + "%6d%9d%8d%9d\n", + obj->name, obj->size, + atomic_read(&obj->handlecount.refcount), + atomic_read(&obj->refcount.refcount)); + if (nid->len > DRM_PROC_LIMIT) { + nid->eof = 1; + return 0; + } + return 0; +} + +static int drm_gem_name_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + struct drm_gem_name_info_data nid; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + nid.len = sprintf(buf, " name size handles refcount\n"); + nid.buf = buf; + nid.eof = 0; + idr_for_each(&dev->object_name_idr, drm_gem_one_name_info, &nid); + + *start = &buf[offset]; + *eof = 0; + if (nid.len > request + offset) + return request; + *eof = 1; + return nid.len - offset; +} + +static int drm_gem_object_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("%d objects\n", atomic_read(&dev->object_count)); + DRM_PROC_PRINT("%d object bytes\n", atomic_read(&dev->object_memory)); + DRM_PROC_PRINT("%d pinned\n", atomic_read(&dev->pin_count)); + DRM_PROC_PRINT("%d pin bytes\n", atomic_read(&dev->pin_memory)); + DRM_PROC_PRINT("%d gtt bytes\n", atomic_read(&dev->gtt_memory)); + DRM_PROC_PRINT("%d gtt total\n", dev->gtt_total); + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + #if DRM_DEBUG_CODE static int drm__vma_info(char *buf, char **start, off_t offset, int request, diff --git a/drivers/gpu/drm/drm_stub.c b/drivers/gpu/drm/drm_stub.c index c2f584f3b46c..82f4657b8879 100644 --- a/drivers/gpu/drm/drm_stub.c +++ b/drivers/gpu/drm/drm_stub.c @@ -152,6 +152,15 @@ static int drm_fill_in_dev(struct drm_device * dev, struct pci_dev *pdev, goto error_out_unreg; } + if (driver->driver_features & DRIVER_GEM) { + retcode = drm_gem_init(dev); + if (retcode) { + DRM_ERROR("Cannot initialize graphics execution " + "manager (GEM)\n"); + goto error_out_unreg; + } + } + return 0; error_out_unreg: @@ -317,6 +326,7 @@ int drm_put_dev(struct drm_device * dev) int drm_put_minor(struct drm_minor **minor_p) { struct drm_minor *minor = *minor_p; + DRM_DEBUG("release secondary minor %d\n", minor->index); if (minor->type == DRM_MINOR_LEGACY) diff --git a/drivers/gpu/drm/i915/Makefile b/drivers/gpu/drm/i915/Makefile index c4bbda662e26..5ba78e4fd2b5 100644 --- a/drivers/gpu/drm/i915/Makefile +++ b/drivers/gpu/drm/i915/Makefile @@ -4,7 +4,11 @@ ccflags-y := -Iinclude/drm i915-y := i915_drv.o i915_dma.o i915_irq.o i915_mem.o i915_opregion.o \ - i915_suspend.o + i915_suspend.o \ + i915_gem.o \ + i915_gem_debug.o \ + i915_gem_proc.o \ + i915_gem_tiling.o i915-$(CONFIG_COMPAT) += i915_ioc32.o diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c index 63c6803d471b..f167ff68a4bf 100644 --- a/drivers/gpu/drm/i915/i915_dma.c +++ b/drivers/gpu/drm/i915/i915_dma.c @@ -170,24 +170,31 @@ static int i915_initialize(struct drm_device * dev, drm_i915_init_t * init) dev_priv->sarea_priv = (drm_i915_sarea_t *) ((u8 *) dev_priv->sarea->handle + init->sarea_priv_offset); - dev_priv->ring.Start = init->ring_start; - dev_priv->ring.End = init->ring_end; - dev_priv->ring.Size = init->ring_size; - dev_priv->ring.tail_mask = dev_priv->ring.Size - 1; + if (init->ring_size != 0) { + if (dev_priv->ring.ring_obj != NULL) { + i915_dma_cleanup(dev); + DRM_ERROR("Client tried to initialize ringbuffer in " + "GEM mode\n"); + return -EINVAL; + } - dev_priv->ring.map.offset = init->ring_start; - dev_priv->ring.map.size = init->ring_size; - dev_priv->ring.map.type = 0; - dev_priv->ring.map.flags = 0; - dev_priv->ring.map.mtrr = 0; + dev_priv->ring.Size = init->ring_size; + dev_priv->ring.tail_mask = dev_priv->ring.Size - 1; - drm_core_ioremap(&dev_priv->ring.map, dev); + dev_priv->ring.map.offset = init->ring_start; + dev_priv->ring.map.size = init->ring_size; + dev_priv->ring.map.type = 0; + dev_priv->ring.map.flags = 0; + dev_priv->ring.map.mtrr = 0; - if (dev_priv->ring.map.handle == NULL) { - i915_dma_cleanup(dev); - DRM_ERROR("can not ioremap virtual address for" - " ring buffer\n"); - return -ENOMEM; + drm_core_ioremap(&dev_priv->ring.map, dev); + + if (dev_priv->ring.map.handle == NULL) { + i915_dma_cleanup(dev); + DRM_ERROR("can not ioremap virtual address for" + " ring buffer\n"); + return -ENOMEM; + } } dev_priv->ring.virtual_start = dev_priv->ring.map.handle; @@ -377,9 +384,10 @@ static int i915_emit_cmds(struct drm_device * dev, int __user * buffer, int dwor return 0; } -static int i915_emit_box(struct drm_device * dev, - struct drm_clip_rect __user * boxes, - int i, int DR1, int DR4) +int +i915_emit_box(struct drm_device *dev, + struct drm_clip_rect __user *boxes, + int i, int DR1, int DR4) { drm_i915_private_t *dev_priv = dev->dev_private; struct drm_clip_rect box; @@ -681,6 +689,9 @@ static int i915_getparam(struct drm_device *dev, void *data, case I915_PARAM_LAST_DISPATCH: value = READ_BREADCRUMB(dev_priv); break; + case I915_PARAM_HAS_GEM: + value = 1; + break; default: DRM_ERROR("Unknown parameter %d\n", param->param); return -EINVAL; @@ -784,6 +795,7 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) memset(dev_priv, 0, sizeof(drm_i915_private_t)); dev->dev_private = (void *)dev_priv; + dev_priv->dev = dev; /* Add register map (needed for suspend/resume) */ base = drm_get_resource_start(dev, mmio_bar); @@ -793,6 +805,8 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) _DRM_KERNEL | _DRM_DRIVER, &dev_priv->mmio_map); + i915_gem_load(dev); + /* Init HWS */ if (!I915_NEED_GFX_HWS(dev)) { ret = i915_init_phys_hws(dev); @@ -838,6 +852,25 @@ int i915_driver_unload(struct drm_device *dev) return 0; } +int i915_driver_open(struct drm_device *dev, struct drm_file *file_priv) +{ + struct drm_i915_file_private *i915_file_priv; + + DRM_DEBUG("\n"); + i915_file_priv = (struct drm_i915_file_private *) + drm_alloc(sizeof(*i915_file_priv), DRM_MEM_FILES); + + if (!i915_file_priv) + return -ENOMEM; + + file_priv->driver_priv = i915_file_priv; + + i915_file_priv->mm.last_gem_seqno = 0; + i915_file_priv->mm.last_gem_throttle_seqno = 0; + + return 0; +} + void i915_driver_lastclose(struct drm_device * dev) { drm_i915_private_t *dev_priv = dev->dev_private; @@ -845,6 +878,8 @@ void i915_driver_lastclose(struct drm_device * dev) if (!dev_priv) return; + i915_gem_lastclose(dev); + if (dev_priv->agp_heap) i915_mem_takedown(&(dev_priv->agp_heap)); @@ -857,6 +892,13 @@ void i915_driver_preclose(struct drm_device * dev, struct drm_file *file_priv) i915_mem_release(dev, file_priv, dev_priv->agp_heap); } +void i915_driver_postclose(struct drm_device *dev, struct drm_file *file_priv) +{ + struct drm_i915_file_private *i915_file_priv = file_priv->driver_priv; + + drm_free(i915_file_priv, sizeof(*i915_file_priv), DRM_MEM_FILES); +} + struct drm_ioctl_desc i915_ioctls[] = { DRM_IOCTL_DEF(DRM_I915_INIT, i915_dma_init, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY), DRM_IOCTL_DEF(DRM_I915_FLUSH, i915_flush_ioctl, DRM_AUTH), @@ -875,6 +917,22 @@ struct drm_ioctl_desc i915_ioctls[] = { DRM_IOCTL_DEF(DRM_I915_GET_VBLANK_PIPE, i915_vblank_pipe_get, DRM_AUTH ), DRM_IOCTL_DEF(DRM_I915_VBLANK_SWAP, i915_vblank_swap, DRM_AUTH), DRM_IOCTL_DEF(DRM_I915_HWS_ADDR, i915_set_status_page, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_INIT, i915_gem_init_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_EXECBUFFER, i915_gem_execbuffer, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_PIN, i915_gem_pin_ioctl, DRM_AUTH|DRM_ROOT_ONLY), + DRM_IOCTL_DEF(DRM_I915_GEM_UNPIN, i915_gem_unpin_ioctl, DRM_AUTH|DRM_ROOT_ONLY), + DRM_IOCTL_DEF(DRM_I915_GEM_BUSY, i915_gem_busy_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_THROTTLE, i915_gem_throttle_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_ENTERVT, i915_gem_entervt_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_LEAVEVT, i915_gem_leavevt_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_CREATE, i915_gem_create_ioctl, 0), + DRM_IOCTL_DEF(DRM_I915_GEM_PREAD, i915_gem_pread_ioctl, 0), + DRM_IOCTL_DEF(DRM_I915_GEM_PWRITE, i915_gem_pwrite_ioctl, 0), + 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); diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 37af03f4db36..a80ead215282 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/drivers/gpu/drm/i915/i915_drv.c @@ -85,12 +85,15 @@ static struct drm_driver driver = { /* don't use mtrr's here, the Xserver or user space app should * deal with them for intel hardware. */ - .driver_features = DRIVER_USE_AGP | DRIVER_REQUIRE_AGP | - DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED, + .driver_features = + DRIVER_USE_AGP | DRIVER_REQUIRE_AGP | /* DRIVER_USE_MTRR |*/ + DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_GEM, .load = i915_driver_load, .unload = i915_driver_unload, + .open = i915_driver_open, .lastclose = i915_driver_lastclose, .preclose = i915_driver_preclose, + .postclose = i915_driver_postclose, .suspend = i915_suspend, .resume = i915_resume, .device_is_agp = i915_driver_device_is_agp, @@ -104,6 +107,10 @@ static struct drm_driver driver = { .reclaim_buffers = drm_core_reclaim_buffers, .get_map_ofs = drm_core_get_map_ofs, .get_reg_ofs = drm_core_get_reg_ofs, + .proc_init = i915_gem_proc_init, + .proc_cleanup = i915_gem_proc_cleanup, + .gem_init_object = i915_gem_init_object, + .gem_free_object = i915_gem_free_object, .ioctls = i915_ioctls, .fops = { .owner = THIS_MODULE, diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h index d1a02bead458..87b071ab8647 100644 --- a/drivers/gpu/drm/i915/i915_drv.h +++ b/drivers/gpu/drm/i915/i915_drv.h @@ -39,7 +39,7 @@ #define DRIVER_NAME "i915" #define DRIVER_DESC "Intel Graphics" -#define DRIVER_DATE "20060119" +#define DRIVER_DATE "20080730" enum pipe { PIPE_A = 0, @@ -60,16 +60,23 @@ enum pipe { #define DRIVER_MINOR 6 #define DRIVER_PATCHLEVEL 0 +#define WATCH_COHERENCY 0 +#define WATCH_BUF 0 +#define WATCH_EXEC 0 +#define WATCH_LRU 0 +#define WATCH_RELOC 0 +#define WATCH_INACTIVE 0 +#define WATCH_PWRITE 0 + typedef struct _drm_i915_ring_buffer { int tail_mask; - unsigned long Start; - unsigned long End; unsigned long Size; u8 *virtual_start; int head; int tail; int space; drm_local_map_t map; + struct drm_gem_object *ring_obj; } drm_i915_ring_buffer_t; struct mem_block { @@ -101,6 +108,8 @@ struct intel_opregion { }; typedef struct drm_i915_private { + struct drm_device *dev; + drm_local_map_t *sarea; drm_local_map_t *mmio_map; @@ -113,6 +122,7 @@ typedef struct drm_i915_private { uint32_t counter; unsigned int status_gfx_addr; drm_local_map_t hws_map; + struct drm_gem_object *hws_obj; unsigned int cpp; int back_offset; @@ -122,7 +132,6 @@ typedef struct drm_i915_private { wait_queue_head_t irq_queue; atomic_t irq_received; - atomic_t irq_emitted; /** Protects user_irq_refcount and irq_mask_reg */ spinlock_t user_irq_lock; /** Refcount for i915_user_irq_get() versus i915_user_irq_put(). */ @@ -230,8 +239,174 @@ typedef struct drm_i915_private { u8 saveDACMASK; u8 saveDACDATA[256*3]; /* 256 3-byte colors */ u8 saveCR[37]; + + struct { + struct drm_mm gtt_space; + + /** + * List of objects currently involved in rendering from the + * ringbuffer. + * + * A reference is held on the buffer while on this list. + */ + struct list_head active_list; + + /** + * List of objects which are not in the ringbuffer but which + * still have a write_domain which needs to be flushed before + * unbinding. + * + * A reference is held on the buffer while on this list. + */ + struct list_head flushing_list; + + /** + * LRU list of objects which are not in the ringbuffer and + * are ready to unbind, but are still in the GTT. + * + * A reference is not held on the buffer while on this list, + * as merely being GTT-bound shouldn't prevent its being + * freed, and we'll pull it off the list in the free path. + */ + struct list_head inactive_list; + + /** + * List of breadcrumbs associated with GPU requests currently + * outstanding. + */ + struct list_head request_list; + + /** + * We leave the user IRQ off as much as possible, + * but this means that requests will finish and never + * be retired once the system goes idle. Set a timer to + * fire periodically while the ring is running. When it + * fires, go retire requests. + */ + struct delayed_work retire_work; + + uint32_t next_gem_seqno; + + /** + * Waiting sequence number, if any + */ + uint32_t waiting_gem_seqno; + + /** + * Last seq seen at irq time + */ + uint32_t irq_gem_seqno; + + /** + * Flag if the X Server, and thus DRM, is not currently in + * control of the device. + * + * This is set between LeaveVT and EnterVT. It needs to be + * replaced with a semaphore. It also needs to be + * transitioned away from for kernel modesetting. + */ + int suspended; + + /** + * Flag if the hardware appears to be wedged. + * + * This is set when attempts to idle the device timeout. + * It prevents command submission from occuring and makes + * 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; +/** driver private structure attached to each drm_gem_object */ +struct drm_i915_gem_object { + struct drm_gem_object *obj; + + /** Current space allocated to this object in the GTT, if any. */ + struct drm_mm_node *gtt_space; + + /** This object's place on the active/flushing/inactive lists */ + struct list_head list; + + /** + * This is set if the object is on the active or flushing lists + * (has pending rendering), and is not set if it's on inactive (ready + * to be unbound). + */ + int active; + + /** + * This is set if the object has been written to since last bound + * to the GTT + */ + int dirty; + + /** AGP memory structure for our GTT binding. */ + DRM_AGP_MEM *agp_mem; + + struct page **page_list; + + /** + * Current offset of the object in GTT space. + * + * This is the same as gtt_space->start + */ + uint32_t gtt_offset; + + /** Boolean whether this object has a valid gtt offset. */ + int gtt_bound; + + /** How many users have pinned this object in GTT space */ + int pin_count; + + /** Breadcrumb of last rendering to the buffer. */ + uint32_t last_rendering_seqno; + + /** Current tiling mode for the object. */ + uint32_t tiling_mode; + + /** + * Flagging of which individual pages are valid in GEM_DOMAIN_CPU when + * GEM_DOMAIN_CPU is not in the object's read domain. + */ + uint8_t *page_cpu_valid; +}; + +/** + * Request queue structure. + * + * The request queue allows us to note sequence numbers that have been emitted + * and may be associated with active buffers to be retired. + * + * By keeping this list, we can avoid having to do questionable + * sequence-number comparisons on buffer last_rendering_seqnos, and associate + * an emission time with seqnos for tracking how far ahead of the GPU we are. + */ +struct drm_i915_gem_request { + /** GEM sequence number associated with this request. */ + uint32_t seqno; + + /** Time at which this request was emitted, in jiffies. */ + unsigned long emitted_jiffies; + + /** Cache domains that were flushed at the start of the request. */ + uint32_t flush_domains; + + struct list_head list; +}; + +struct drm_i915_file_private { + struct { + uint32_t last_gem_seqno; + uint32_t last_gem_throttle_seqno; + } mm; +}; + extern struct drm_ioctl_desc i915_ioctls[]; extern int i915_max_ioctl; @@ -239,18 +414,26 @@ extern int i915_max_ioctl; extern void i915_kernel_lost_context(struct drm_device * dev); extern int i915_driver_load(struct drm_device *, unsigned long flags); extern int i915_driver_unload(struct drm_device *); +extern int i915_driver_open(struct drm_device *dev, struct drm_file *file_priv); extern void i915_driver_lastclose(struct drm_device * dev); extern void i915_driver_preclose(struct drm_device *dev, struct drm_file *file_priv); +extern void i915_driver_postclose(struct drm_device *dev, + struct drm_file *file_priv); extern int i915_driver_device_is_agp(struct drm_device * dev); extern long i915_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg); +extern int i915_emit_box(struct drm_device *dev, + struct drm_clip_rect __user *boxes, + int i, int DR1, int DR4); /* i915_irq.c */ extern int i915_irq_emit(struct drm_device *dev, void *data, struct drm_file *file_priv); extern int i915_irq_wait(struct drm_device *dev, void *data, struct drm_file *file_priv); +void i915_user_irq_get(struct drm_device *dev); +void i915_user_irq_put(struct drm_device *dev); extern irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS); extern void i915_driver_irq_preinstall(struct drm_device * dev); @@ -279,6 +462,67 @@ extern int i915_mem_destroy_heap(struct drm_device *dev, void *data, extern void i915_mem_takedown(struct mem_block **heap); extern void i915_mem_release(struct drm_device * dev, struct drm_file *file_priv, struct mem_block *heap); +/* i915_gem.c */ +int i915_gem_init_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_create_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_pread_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_mmap_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_execbuffer(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_pin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_unpin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_busy_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_throttle_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +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); +void i915_gem_free_object(struct drm_gem_object *obj); +int i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment); +void i915_gem_object_unpin(struct drm_gem_object *obj); +void i915_gem_lastclose(struct drm_device *dev); +uint32_t i915_get_gem_seqno(struct drm_device *dev); +void i915_gem_retire_requests(struct drm_device *dev); +void i915_gem_retire_work_handler(struct work_struct *work); +void i915_gem_clflush_object(struct drm_gem_object *obj); + +/* i915_gem_tiling.c */ +void i915_gem_detect_bit_6_swizzle(struct drm_device *dev); + +/* i915_gem_debug.c */ +void i915_gem_dump_object(struct drm_gem_object *obj, int len, + const char *where, uint32_t mark); +#if WATCH_INACTIVE +void i915_verify_inactive(struct drm_device *dev, char *file, int line); +#else +#define i915_verify_inactive(dev, file, line) +#endif +void i915_gem_object_check_coherency(struct drm_gem_object *obj, int handle); +void i915_gem_dump_object(struct drm_gem_object *obj, int len, + const char *where, uint32_t mark); +void i915_dump_lru(struct drm_device *dev, const char *where); /* i915_suspend.c */ extern int i915_save_state(struct drm_device *dev); @@ -347,6 +591,7 @@ extern void opregion_enable_asle(struct drm_device *dev); */ #define READ_HWSP(dev_priv, reg) (((volatile u32*)(dev_priv->hw_status_page))[reg]) #define READ_BREADCRUMB(dev_priv) READ_HWSP(dev_priv, 5) +#define I915_GEM_HWS_INDEX 0x10 extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller); diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c new file mode 100644 index 000000000000..90ae8a0369f7 --- /dev/null +++ b/drivers/gpu/drm/i915/i915_gem.c @@ -0,0 +1,2497 @@ +/* + * 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" +#include <linux/swap.h> + +static int +i915_gem_object_set_domain(struct drm_gem_object *obj, + uint32_t read_domains, + uint32_t write_domain); +static int +i915_gem_object_set_domain_range(struct drm_gem_object *obj, + uint64_t offset, + uint64_t size, + uint32_t read_domains, + uint32_t write_domain); +static int +i915_gem_set_domain(struct drm_gem_object *obj, + struct drm_file *file_priv, + uint32_t read_domains, + uint32_t write_domain); +static int i915_gem_object_get_page_list(struct drm_gem_object *obj); +static void i915_gem_object_free_page_list(struct drm_gem_object *obj); +static int i915_gem_object_wait_rendering(struct drm_gem_object *obj); + +int +i915_gem_init_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_init *args = data; + + mutex_lock(&dev->struct_mutex); + + if (args->gtt_start >= args->gtt_end || + (args->gtt_start & (PAGE_SIZE - 1)) != 0 || + (args->gtt_end & (PAGE_SIZE - 1)) != 0) { + mutex_unlock(&dev->struct_mutex); + return -EINVAL; + } + + drm_mm_init(&dev_priv->mm.gtt_space, args->gtt_start, + args->gtt_end - args->gtt_start); + + dev->gtt_total = (uint32_t) (args->gtt_end - args->gtt_start); + + mutex_unlock(&dev->struct_mutex); + + return 0; +} + + +/** + * Creates a new mm object and returns a handle to it. + */ +int +i915_gem_create_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_create *args = data; + struct drm_gem_object *obj; + int handle, ret; + + args->size = roundup(args->size, PAGE_SIZE); + + /* Allocate the new object */ + obj = drm_gem_object_alloc(dev, args->size); + if (obj == NULL) + return -ENOMEM; + + ret = drm_gem_handle_create(file_priv, obj, &handle); + mutex_lock(&dev->struct_mutex); + drm_gem_object_handle_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + if (ret) + return ret; + + args->handle = handle; + + return 0; +} + +/** + * Reads data from the object referenced by handle. + * + * On error, the contents of *data are undefined. + */ +int +i915_gem_pread_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_pread *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + ssize_t read; + loff_t offset; + int ret; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EBADF; + obj_priv = obj->driver_private; + + /* Bounds check source. + * + * XXX: This could use review for overflow issues... + */ + if (args->offset > obj->size || args->size > obj->size || + args->offset + args->size > obj->size) { + drm_gem_object_unreference(obj); + return -EINVAL; + } + + mutex_lock(&dev->struct_mutex); + + ret = i915_gem_object_set_domain_range(obj, args->offset, args->size, + I915_GEM_DOMAIN_CPU, 0); + if (ret != 0) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + } + + offset = args->offset; + + read = vfs_read(obj->filp, (char __user *)(uintptr_t)args->data_ptr, + args->size, &offset); + if (read != args->size) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (read < 0) + return read; + else + return -EINVAL; + } + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +static int +i915_gem_gtt_pwrite(struct drm_device *dev, struct drm_gem_object *obj, + struct drm_i915_gem_pwrite *args, + struct drm_file *file_priv) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + ssize_t remain; + loff_t offset; + char __user *user_data; + char *vaddr; + int i, o, l; + int ret = 0; + unsigned long pfn; + unsigned long unwritten; + + user_data = (char __user *) (uintptr_t) args->data_ptr; + remain = args->size; + if (!access_ok(VERIFY_READ, user_data, remain)) + return -EFAULT; + + + mutex_lock(&dev->struct_mutex); + ret = i915_gem_object_pin(obj, 0); + if (ret) { + mutex_unlock(&dev->struct_mutex); + return ret; + } + ret = i915_gem_set_domain(obj, file_priv, + I915_GEM_DOMAIN_GTT, I915_GEM_DOMAIN_GTT); + if (ret) + goto fail; + + obj_priv = obj->driver_private; + offset = obj_priv->gtt_offset + args->offset; + obj_priv->dirty = 1; + + while (remain > 0) { + /* Operation in this page + * + * i = page number + * o = offset within page + * l = bytes to copy + */ + i = offset >> PAGE_SHIFT; + o = offset & (PAGE_SIZE-1); + l = remain; + if ((o + l) > PAGE_SIZE) + l = PAGE_SIZE - o; + + pfn = (dev->agp->base >> PAGE_SHIFT) + i; + +#ifdef CONFIG_HIGHMEM + /* kmap_atomic can't map IO pages on non-HIGHMEM kernels + */ + vaddr = kmap_atomic_pfn(pfn, KM_USER0); +#if WATCH_PWRITE + DRM_INFO("pwrite i %d o %d l %d pfn %ld vaddr %p\n", + i, o, l, pfn, vaddr); +#endif + unwritten = __copy_from_user_inatomic_nocache(vaddr + o, + user_data, l); + kunmap_atomic(vaddr, KM_USER0); + + if (unwritten) +#endif /* CONFIG_HIGHMEM */ + { + vaddr = ioremap(pfn << PAGE_SHIFT, PAGE_SIZE); +#if WATCH_PWRITE + DRM_INFO("pwrite slow i %d o %d l %d " + "pfn %ld vaddr %p\n", + i, o, l, pfn, vaddr); +#endif + if (vaddr == NULL) { + ret = -EFAULT; + goto fail; + } + unwritten = __copy_from_user(vaddr + o, user_data, l); +#if WATCH_PWRITE + DRM_INFO("unwritten %ld\n", unwritten); +#endif + iounmap(vaddr); + if (unwritten) { + ret = -EFAULT; + goto fail; + } + } + + remain -= l; + user_data += l; + offset += l; + } +#if WATCH_PWRITE && 1 + i915_gem_clflush_object(obj); + i915_gem_dump_object(obj, args->offset + args->size, __func__, ~0); + i915_gem_clflush_object(obj); +#endif + +fail: + i915_gem_object_unpin(obj); + mutex_unlock(&dev->struct_mutex); + + return ret; +} + +int +i915_gem_shmem_pwrite(struct drm_device *dev, struct drm_gem_object *obj, + struct drm_i915_gem_pwrite *args, + struct drm_file *file_priv) +{ + int ret; + loff_t offset; + ssize_t written; + + mutex_lock(&dev->struct_mutex); + + ret = i915_gem_set_domain(obj, file_priv, + I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU); + if (ret) { + mutex_unlock(&dev->struct_mutex); + return ret; + } + + offset = args->offset; + + written = vfs_write(obj->filp, + (char __user *)(uintptr_t) args->data_ptr, + args->size, &offset); + if (written != args->size) { + mutex_unlock(&dev->struct_mutex); + if (written < 0) + return written; + else + return -EINVAL; + } + + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +/** + * Writes data to the object referenced by handle. + * + * On error, the contents of the buffer that were to be modified are undefined. + */ +int +i915_gem_pwrite_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_pwrite *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret = 0; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EBADF; + obj_priv = obj->driver_private; + + /* Bounds check destination. + * + * XXX: This could use review for overflow issues... + */ + if (args->offset > obj->size || args->size > obj->size || + args->offset + args->size > obj->size) { + drm_gem_object_unreference(obj); + return -EINVAL; + } + + /* We can only do the GTT pwrite on untiled buffers, as otherwise + * it would end up going through the fenced access, and we'll get + * different detiling behavior between reading and writing. + * pread/pwrite currently are reading and writing from the CPU + * perspective, requiring manual detiling by the client. + */ + if (obj_priv->tiling_mode == I915_TILING_NONE && + dev->gtt_total != 0) + ret = i915_gem_gtt_pwrite(dev, obj, args, file_priv); + else + ret = i915_gem_shmem_pwrite(dev, obj, args, file_priv); + +#if WATCH_PWRITE + if (ret) + DRM_INFO("pwrite failed %d\n", ret); +#endif + + drm_gem_object_unreference(obj); + + return ret; +} + +/** + * Called when user space prepares to use an object + */ +int +i915_gem_set_domain_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_set_domain *args = data; + struct drm_gem_object *obj; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EBADF; + + mutex_lock(&dev->struct_mutex); +#if WATCH_BUF + DRM_INFO("set_domain_ioctl %p(%d), %08x %08x\n", + obj, obj->size, args->read_domains, args->write_domain); +#endif + ret = i915_gem_set_domain(obj, file_priv, + args->read_domains, args->write_domain); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; +} + +/** + * Called when user space has done writes to this buffer + */ +int +i915_gem_sw_finish_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_sw_finish *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret = 0; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + mutex_lock(&dev->struct_mutex); + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + mutex_unlock(&dev->struct_mutex); + return -EBADF; + } + +#if WATCH_BUF + DRM_INFO("%s: sw_finish %d (%p %d)\n", + __func__, args->handle, obj, obj->size); +#endif + obj_priv = obj->driver_private; + + /* Pinned buffers may be scanout, so flush the cache */ + if ((obj->write_domain & I915_GEM_DOMAIN_CPU) && obj_priv->pin_count) { + i915_gem_clflush_object(obj); + drm_agp_chipset_flush(dev); + } + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; +} + +/** + * Maps the contents of an object, returning the address it is mapped + * into. + * + * While the mapping holds a reference on the contents of the object, it doesn't + * imply a ref on the object itself. + */ +int +i915_gem_mmap_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_mmap *args = data; + struct drm_gem_object *obj; + loff_t offset; + unsigned long addr; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EBADF; + + offset = args->offset; + + down_write(¤t->mm->mmap_sem); + addr = do_mmap(obj->filp, 0, args->size, + PROT_READ | PROT_WRITE, MAP_SHARED, + args->offset); + up_write(¤t->mm->mmap_sem); + mutex_lock(&dev->struct_mutex); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (IS_ERR((void *)addr)) + return addr; + + args->addr_ptr = (uint64_t) addr; + + return 0; +} + +static void +i915_gem_object_free_page_list(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page_count = obj->size / PAGE_SIZE; + int i; + + if (obj_priv->page_list == NULL) + return; + + + for (i = 0; i < page_count; i++) + if (obj_priv->page_list[i] != NULL) { + if (obj_priv->dirty) + set_page_dirty(obj_priv->page_list[i]); + mark_page_accessed(obj_priv->page_list[i]); + page_cache_release(obj_priv->page_list[i]); + } + obj_priv->dirty = 0; + + drm_free(obj_priv->page_list, + page_count * sizeof(struct page *), + DRM_MEM_DRIVER); + obj_priv->page_list = NULL; +} + +static void +i915_gem_object_move_to_active(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + /* Add a reference if we're newly entering the active list. */ + if (!obj_priv->active) { + drm_gem_object_reference(obj); + obj_priv->active = 1; + } + /* Move from whatever list we were on to the tail of execution. */ + list_move_tail(&obj_priv->list, + &dev_priv->mm.active_list); +} + + +static void +i915_gem_object_move_to_inactive(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + i915_verify_inactive(dev, __FILE__, __LINE__); + if (obj_priv->pin_count != 0) + list_del_init(&obj_priv->list); + else + list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list); + + if (obj_priv->active) { + obj_priv->active = 0; + drm_gem_object_unreference(obj); + } + i915_verify_inactive(dev, __FILE__, __LINE__); +} + +/** + * Creates a new sequence number, emitting a write of it to the status page + * plus an interrupt, which will trigger i915_user_interrupt_handler. + * + * Must be called with struct_lock held. + * + * Returned sequence numbers are nonzero on success. + */ +static uint32_t +i915_add_request(struct drm_device *dev, uint32_t flush_domains) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_request *request; + uint32_t seqno; + int was_empty; + RING_LOCALS; + + request = drm_calloc(1, sizeof(*request), DRM_MEM_DRIVER); + if (request == NULL) + return 0; + + /* Grab the seqno we're going to make this request be, and bump the + * next (skipping 0 so it can be the reserved no-seqno value). + */ + seqno = dev_priv->mm.next_gem_seqno; + dev_priv->mm.next_gem_seqno++; + if (dev_priv->mm.next_gem_seqno == 0) + dev_priv->mm.next_gem_seqno++; + + BEGIN_LP_RING(4); + OUT_RING(MI_STORE_DWORD_INDEX); + OUT_RING(I915_GEM_HWS_INDEX << MI_STORE_DWORD_INDEX_SHIFT); + OUT_RING(seqno); + + OUT_RING(MI_USER_INTERRUPT); + ADVANCE_LP_RING(); + + DRM_DEBUG("%d\n", seqno); + + request->seqno = seqno; + request->emitted_jiffies = jiffies; + request->flush_domains = flush_domains; + was_empty = list_empty(&dev_priv->mm.request_list); + list_add_tail(&request->list, &dev_priv->mm.request_list); + + if (was_empty) + schedule_delayed_work(&dev_priv->mm.retire_work, HZ); + return seqno; +} + +/** + * Command execution barrier + * + * Ensures that all commands in the ring are finished + * before signalling the CPU + */ +uint32_t +i915_retire_commands(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t cmd = MI_FLUSH | MI_NO_WRITE_FLUSH; + uint32_t flush_domains = 0; + RING_LOCALS; + + /* The sampler always gets flushed on i965 (sigh) */ + if (IS_I965G(dev)) + flush_domains |= I915_GEM_DOMAIN_SAMPLER; + BEGIN_LP_RING(2); + OUT_RING(cmd); + OUT_RING(0); /* noop */ + ADVANCE_LP_RING(); + return flush_domains; +} + +/** + * Moves buffers associated only with the given active seqno from the active + * to inactive list, potentially freeing them. + */ +static void +i915_gem_retire_request(struct drm_device *dev, + struct drm_i915_gem_request *request) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + + /* Move any buffers on the active list that are no longer referenced + * by the ringbuffer to the flushing/inactive lists as appropriate. + */ + while (!list_empty(&dev_priv->mm.active_list)) { + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + obj_priv = list_first_entry(&dev_priv->mm.active_list, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + + /* If the seqno being retired doesn't match the oldest in the + * list, then the oldest in the list must still be newer than + * this seqno. + */ + if (obj_priv->last_rendering_seqno != request->seqno) + return; +#if WATCH_LRU + DRM_INFO("%s: retire %d moves to inactive list %p\n", + __func__, request->seqno, obj); +#endif + + if (obj->write_domain != 0) { + list_move_tail(&obj_priv->list, + &dev_priv->mm.flushing_list); + } else { + i915_gem_object_move_to_inactive(obj); + } + } + + if (request->flush_domains != 0) { + struct drm_i915_gem_object *obj_priv, *next; + + /* Clear the write domain and activity from any buffers + * that are just waiting for a flush matching the one retired. + */ + list_for_each_entry_safe(obj_priv, next, + &dev_priv->mm.flushing_list, list) { + struct drm_gem_object *obj = obj_priv->obj; + + if (obj->write_domain & request->flush_domains) { + obj->write_domain = 0; + i915_gem_object_move_to_inactive(obj); + } + } + + } +} + +/** + * Returns true if seq1 is later than seq2. + */ +static int +i915_seqno_passed(uint32_t seq1, uint32_t seq2) +{ + return (int32_t)(seq1 - seq2) >= 0; +} + +uint32_t +i915_get_gem_seqno(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + + return READ_HWSP(dev_priv, I915_GEM_HWS_INDEX); +} + +/** + * This function clears the request list as sequence numbers are passed. + */ +void +i915_gem_retire_requests(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t seqno; + + seqno = i915_get_gem_seqno(dev); + + while (!list_empty(&dev_priv->mm.request_list)) { + struct drm_i915_gem_request *request; + uint32_t retiring_seqno; + + request = list_first_entry(&dev_priv->mm.request_list, + struct drm_i915_gem_request, + list); + retiring_seqno = request->seqno; + + if (i915_seqno_passed(seqno, retiring_seqno) || + dev_priv->mm.wedged) { + i915_gem_retire_request(dev, request); + + list_del(&request->list); + drm_free(request, sizeof(*request), DRM_MEM_DRIVER); + } else + break; + } +} + +void +i915_gem_retire_work_handler(struct work_struct *work) +{ + drm_i915_private_t *dev_priv; + struct drm_device *dev; + + dev_priv = container_of(work, drm_i915_private_t, + mm.retire_work.work); + dev = dev_priv->dev; + + mutex_lock(&dev->struct_mutex); + i915_gem_retire_requests(dev); + if (!list_empty(&dev_priv->mm.request_list)) + schedule_delayed_work(&dev_priv->mm.retire_work, HZ); + mutex_unlock(&dev->struct_mutex); +} + +/** + * Waits for a sequence number to be signaled, and cleans up the + * request and object lists appropriately for that event. + */ +int +i915_wait_request(struct drm_device *dev, uint32_t seqno) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + int ret = 0; + + BUG_ON(seqno == 0); + + if (!i915_seqno_passed(i915_get_gem_seqno(dev), seqno)) { + dev_priv->mm.waiting_gem_seqno = seqno; + i915_user_irq_get(dev); + ret = wait_event_interruptible(dev_priv->irq_queue, + i915_seqno_passed(i915_get_gem_seqno(dev), + seqno) || + dev_priv->mm.wedged); + i915_user_irq_put(dev); + dev_priv->mm.waiting_gem_seqno = 0; + } + if (dev_priv->mm.wedged) + ret = -EIO; + + if (ret && ret != -ERESTARTSYS) + DRM_ERROR("%s returns %d (awaiting %d at %d)\n", + __func__, ret, seqno, i915_get_gem_seqno(dev)); + + /* Directly dispatch request retiring. While we have the work queue + * to handle this, the waiter on a request often wants an associated + * buffer to have made it to the inactive list, and we would need + * a separate wait queue to handle that. + */ + if (ret == 0) + i915_gem_retire_requests(dev); + + return ret; +} + +static void +i915_gem_flush(struct drm_device *dev, + uint32_t invalidate_domains, + uint32_t flush_domains) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t cmd; + RING_LOCALS; + +#if WATCH_EXEC + DRM_INFO("%s: invalidate %08x flush %08x\n", __func__, + invalidate_domains, flush_domains); +#endif + + if (flush_domains & I915_GEM_DOMAIN_CPU) + drm_agp_chipset_flush(dev); + + if ((invalidate_domains | flush_domains) & ~(I915_GEM_DOMAIN_CPU | + I915_GEM_DOMAIN_GTT)) { + /* + * read/write caches: + * + * I915_GEM_DOMAIN_RENDER is always invalidated, but is + * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is + * also flushed at 2d versus 3d pipeline switches. + * + * read-only caches: + * + * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if + * MI_READ_FLUSH is set, and is always flushed on 965. + * + * I915_GEM_DOMAIN_COMMAND may not exist? + * + * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is + * invalidated when MI_EXE_FLUSH is set. + * + * I915_GEM_DOMAIN_VERTEX, which exists on 965, is + * invalidated with every MI_FLUSH. + * + * TLBs: + * + * On 965, TLBs associated with I915_GEM_DOMAIN_COMMAND + * and I915_GEM_DOMAIN_CPU in are invalidated at PTE write and + * I915_GEM_DOMAIN_RENDER and I915_GEM_DOMAIN_SAMPLER + * are flushed at any MI_FLUSH. + */ + + cmd = MI_FLUSH | MI_NO_WRITE_FLUSH; + if ((invalidate_domains|flush_domains) & + I915_GEM_DOMAIN_RENDER) + cmd &= ~MI_NO_WRITE_FLUSH; + if (!IS_I965G(dev)) { + /* + * On the 965, the sampler cache always gets flushed + * and this bit is reserved. + */ + if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER) + cmd |= MI_READ_FLUSH; + } + if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION) + cmd |= MI_EXE_FLUSH; + +#if WATCH_EXEC + DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd); +#endif + BEGIN_LP_RING(2); + OUT_RING(cmd); + OUT_RING(0); /* noop */ + ADVANCE_LP_RING(); + } +} + +/** + * Ensures that all rendering to the object has completed and the object is + * safe to unbind from the GTT or access from the CPU. + */ +static int +i915_gem_object_wait_rendering(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int ret; + + /* If there are writes queued to the buffer, flush and + * create a new seqno to wait for. + */ + if (obj->write_domain & ~(I915_GEM_DOMAIN_CPU|I915_GEM_DOMAIN_GTT)) { + uint32_t write_domain = obj->write_domain; +#if WATCH_BUF + DRM_INFO("%s: flushing object %p from write domain %08x\n", + __func__, obj, write_domain); +#endif + i915_gem_flush(dev, 0, write_domain); + + i915_gem_object_move_to_active(obj); + obj_priv->last_rendering_seqno = i915_add_request(dev, + write_domain); + BUG_ON(obj_priv->last_rendering_seqno == 0); +#if WATCH_LRU + DRM_INFO("%s: flush moves to exec list %p\n", __func__, obj); +#endif + } + + /* If there is rendering queued on the buffer being evicted, wait for + * it. + */ + if (obj_priv->active) { +#if WATCH_BUF + DRM_INFO("%s: object %p wait for seqno %08x\n", + __func__, obj, obj_priv->last_rendering_seqno); +#endif + ret = i915_wait_request(dev, obj_priv->last_rendering_seqno); + if (ret != 0) + return ret; + } + + return 0; +} + +/** + * Unbinds an object from the GTT aperture. + */ +static int +i915_gem_object_unbind(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int ret = 0; + +#if WATCH_BUF + DRM_INFO("%s:%d %p\n", __func__, __LINE__, obj); + DRM_INFO("gtt_space %p\n", obj_priv->gtt_space); +#endif + if (obj_priv->gtt_space == NULL) + return 0; + + if (obj_priv->pin_count != 0) { + DRM_ERROR("Attempting to unbind pinned buffer\n"); + return -EINVAL; + } + + /* Wait for any rendering to complete + */ + ret = i915_gem_object_wait_rendering(obj); + if (ret) { + DRM_ERROR("wait_rendering failed: %d\n", ret); + return ret; + } + + /* Move the object to the CPU domain to ensure that + * any possible CPU writes while it's not in the GTT + * are flushed when we go to remap it. This will + * also ensure that all pending GPU writes are finished + * before we unbind. + */ + ret = i915_gem_object_set_domain(obj, I915_GEM_DOMAIN_CPU, + I915_GEM_DOMAIN_CPU); + if (ret) { + DRM_ERROR("set_domain failed: %d\n", ret); + return ret; + } + + if (obj_priv->agp_mem != NULL) { + drm_unbind_agp(obj_priv->agp_mem); + drm_free_agp(obj_priv->agp_mem, obj->size / PAGE_SIZE); + obj_priv->agp_mem = NULL; + } + + BUG_ON(obj_priv->active); + + i915_gem_object_free_page_list(obj); + + if (obj_priv->gtt_space) { + atomic_dec(&dev->gtt_count); + atomic_sub(obj->size, &dev->gtt_memory); + + drm_mm_put_block(obj_priv->gtt_space); + obj_priv->gtt_space = NULL; + } + + /* Remove ourselves from the LRU list if present. */ + if (!list_empty(&obj_priv->list)) + list_del_init(&obj_priv->list); + + return 0; +} + +static int +i915_gem_evict_something(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret = 0; + + for (;;) { + /* If there's an inactive buffer available now, grab it + * and be done. + */ + if (!list_empty(&dev_priv->mm.inactive_list)) { + obj_priv = list_first_entry(&dev_priv->mm.inactive_list, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + BUG_ON(obj_priv->pin_count != 0); +#if WATCH_LRU + DRM_INFO("%s: evicting %p\n", __func__, obj); +#endif + BUG_ON(obj_priv->active); + + /* Wait on the rendering and unbind the buffer. */ + ret = i915_gem_object_unbind(obj); + break; + } + + /* If we didn't get anything, but the ring is still processing + * things, wait for one of those things to finish and hopefully + * leave us a buffer to evict. + */ + if (!list_empty(&dev_priv->mm.request_list)) { + struct drm_i915_gem_request *request; + + request = list_first_entry(&dev_priv->mm.request_list, + struct drm_i915_gem_request, + list); + + ret = i915_wait_request(dev, request->seqno); + if (ret) + break; + + /* if waiting caused an object to become inactive, + * then loop around and wait for it. Otherwise, we + * assume that waiting freed and unbound something, + * so there should now be some space in the GTT + */ + if (!list_empty(&dev_priv->mm.inactive_list)) + continue; + break; + } + + /* If we didn't have anything on the request list but there + * are buffers awaiting a flush, emit one and try again. + * When we wait on it, those buffers waiting for that flush + * will get moved to inactive. + */ + if (!list_empty(&dev_priv->mm.flushing_list)) { + obj_priv = list_first_entry(&dev_priv->mm.flushing_list, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + + i915_gem_flush(dev, + obj->write_domain, + obj->write_domain); + i915_add_request(dev, obj->write_domain); + + obj = NULL; + continue; + } + + DRM_ERROR("inactive empty %d request empty %d " + "flushing empty %d\n", + list_empty(&dev_priv->mm.inactive_list), + list_empty(&dev_priv->mm.request_list), + list_empty(&dev_priv->mm.flushing_list)); + /* If we didn't do any of the above, there's nothing to be done + * and we just can't fit it in. + */ + return -ENOMEM; + } + return ret; +} + +static int +i915_gem_object_get_page_list(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page_count, i; + struct address_space *mapping; + struct inode *inode; + struct page *page; + int ret; + + if (obj_priv->page_list) + return 0; + + /* Get the list of pages out of our struct file. They'll be pinned + * at this point until we release them. + */ + page_count = obj->size / PAGE_SIZE; + BUG_ON(obj_priv->page_list != NULL); + obj_priv->page_list = drm_calloc(page_count, sizeof(struct page *), + DRM_MEM_DRIVER); + if (obj_priv->page_list == NULL) { + DRM_ERROR("Faled to allocate page list\n"); + return -ENOMEM; + } + + inode = obj->filp->f_path.dentry->d_inode; + mapping = inode->i_mapping; + for (i = 0; i < page_count; i++) { + page = read_mapping_page(mapping, i, NULL); + if (IS_ERR(page)) { + ret = PTR_ERR(page); + DRM_ERROR("read_mapping_page failed: %d\n", ret); + i915_gem_object_free_page_list(obj); + return ret; + } + obj_priv->page_list[i] = page; + } + return 0; +} + +/** + * Finds free space in the GTT aperture and binds the object there. + */ +static int +i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + struct drm_mm_node *free_space; + int page_count, ret; + + if (alignment == 0) + alignment = PAGE_SIZE; + if (alignment & (PAGE_SIZE - 1)) { + DRM_ERROR("Invalid object alignment requested %u\n", alignment); + return -EINVAL; + } + + search_free: + free_space = drm_mm_search_free(&dev_priv->mm.gtt_space, + obj->size, alignment, 0); + if (free_space != NULL) { + obj_priv->gtt_space = drm_mm_get_block(free_space, obj->size, + alignment); + if (obj_priv->gtt_space != NULL) { + obj_priv->gtt_space->private = obj; + obj_priv->gtt_offset = obj_priv->gtt_space->start; + } + } + if (obj_priv->gtt_space == NULL) { + /* If the gtt is empty and we're still having trouble + * fitting our object in, we're out of memory. + */ +#if WATCH_LRU + DRM_INFO("%s: GTT full, evicting something\n", __func__); +#endif + if (list_empty(&dev_priv->mm.inactive_list) && + list_empty(&dev_priv->mm.flushing_list) && + list_empty(&dev_priv->mm.active_list)) { + DRM_ERROR("GTT full, but LRU list empty\n"); + return -ENOMEM; + } + + ret = i915_gem_evict_something(dev); + if (ret != 0) { + DRM_ERROR("Failed to evict a buffer %d\n", ret); + return ret; + } + goto search_free; + } + +#if WATCH_BUF + DRM_INFO("Binding object of size %d at 0x%08x\n", + obj->size, obj_priv->gtt_offset); +#endif + ret = i915_gem_object_get_page_list(obj); + if (ret) { + drm_mm_put_block(obj_priv->gtt_space); + obj_priv->gtt_space = NULL; + return ret; + } + + page_count = obj->size / PAGE_SIZE; + /* Create an AGP memory structure pointing at our pages, and bind it + * into the GTT. + */ + obj_priv->agp_mem = drm_agp_bind_pages(dev, + obj_priv->page_list, + page_count, + obj_priv->gtt_offset); + if (obj_priv->agp_mem == NULL) { + i915_gem_object_free_page_list(obj); + drm_mm_put_block(obj_priv->gtt_space); + obj_priv->gtt_space = NULL; + return -ENOMEM; + } + atomic_inc(&dev->gtt_count); + atomic_add(obj->size, &dev->gtt_memory); + + /* Assert that the object is not currently in any GPU domain. As it + * wasn't in the GTT, there shouldn't be any way it could have been in + * a GPU cache + */ + BUG_ON(obj->read_domains & ~(I915_GEM_DOMAIN_CPU|I915_GEM_DOMAIN_GTT)); + BUG_ON(obj->write_domain & ~(I915_GEM_DOMAIN_CPU|I915_GEM_DOMAIN_GTT)); + + return 0; +} + +void +i915_gem_clflush_object(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + /* If we don't have a page list set up, then we're not pinned + * to GPU, and we can ignore the cache flush because it'll happen + * again at bind time. + */ + if (obj_priv->page_list == NULL) + return; + + drm_clflush_pages(obj_priv->page_list, obj->size / PAGE_SIZE); +} + +/* + * Set the next domain for the specified object. This + * may not actually perform the necessary flushing/invaliding though, + * as that may want to be batched with other set_domain operations + * + * This is (we hope) the only really tricky part of gem. The goal + * is fairly simple -- track which caches hold bits of the object + * and make sure they remain coherent. A few concrete examples may + * help to explain how it works. For shorthand, we use the notation + * (read_domains, write_domain), e.g. (CPU, CPU) to indicate the + * a pair of read and write domain masks. + * + * Case 1: the batch buffer + * + * 1. Allocated + * 2. Written by CPU + * 3. Mapped to GTT + * 4. Read by GPU + * 5. Unmapped from GTT + * 6. Freed + * + * Let's take these a step at a time + * + * 1. Allocated + * Pages allocated from the kernel may still have + * cache contents, so we set them to (CPU, CPU) always. + * 2. Written by CPU (using pwrite) + * The pwrite function calls set_domain (CPU, CPU) and + * this function does nothing (as nothing changes) + * 3. Mapped by GTT + * This function asserts that the object is not + * currently in any GPU-based read or write domains + * 4. Read by GPU + * i915_gem_execbuffer calls set_domain (COMMAND, 0). + * As write_domain is zero, this function adds in the + * current read domains (CPU+COMMAND, 0). + * flush_domains is set to CPU. + * invalidate_domains is set to COMMAND + * clflush is run to get data out of the CPU caches + * then i915_dev_set_domain calls i915_gem_flush to + * emit an MI_FLUSH and drm_agp_chipset_flush + * 5. Unmapped from GTT + * i915_gem_object_unbind calls set_domain (CPU, CPU) + * flush_domains and invalidate_domains end up both zero + * so no flushing/invalidating happens + * 6. Freed + * yay, done + * + * Case 2: The shared render buffer + * + * 1. Allocated + * 2. Mapped to GTT + * 3. Read/written by GPU + * 4. set_domain to (CPU,CPU) + * 5. Read/written by CPU + * 6. Read/written by GPU + * + * 1. Allocated + * Same as last example, (CPU, CPU) + * 2. Mapped to GTT + * Nothing changes (assertions find that it is not in the GPU) + * 3. Read/written by GPU + * execbuffer calls set_domain (RENDER, RENDER) + * flush_domains gets CPU + * invalidate_domains gets GPU + * clflush (obj) + * MI_FLUSH and drm_agp_chipset_flush + * 4. set_domain (CPU, CPU) + * flush_domains gets GPU + * invalidate_domains gets CPU + * wait_rendering (obj) to make sure all drawing is complete. + * This will include an MI_FLUSH to get the data from GPU + * to memory + * clflush (obj) to invalidate the CPU cache + * Another MI_FLUSH in i915_gem_flush (eliminate this somehow?) + * 5. Read/written by CPU + * cache lines are loaded and dirtied + * 6. Read written by GPU + * Same as last GPU access + * + * Case 3: The constant buffer + * + * 1. Allocated + * 2. Written by CPU + * 3. Read by GPU + * 4. Updated (written) by CPU again + * 5. Read by GPU + * + * 1. Allocated + * (CPU, CPU) + * 2. Written by CPU + * (CPU, CPU) + * 3. Read by GPU + * (CPU+RENDER, 0) + * flush_domains = CPU + * invalidate_domains = RENDER + * clflush (obj) + * MI_FLUSH + * drm_agp_chipset_flush + * 4. Updated (written) by CPU again + * (CPU, CPU) + * flush_domains = 0 (no previous write domain) + * invalidate_domains = 0 (no new read domains) + * 5. Read by GPU + * (CPU+RENDER, 0) + * flush_domains = CPU + * invalidate_domains = RENDER + * clflush (obj) + * MI_FLUSH + * drm_agp_chipset_flush + */ +static int +i915_gem_object_set_domain(struct drm_gem_object *obj, + uint32_t read_domains, + uint32_t write_domain) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + uint32_t invalidate_domains = 0; + uint32_t flush_domains = 0; + int ret; + +#if WATCH_BUF + DRM_INFO("%s: object %p read %08x -> %08x write %08x -> %08x\n", + __func__, obj, + obj->read_domains, read_domains, + obj->write_domain, write_domain); +#endif + /* + * If the object isn't moving to a new write domain, + * let the object stay in multiple read domains + */ + if (write_domain == 0) + read_domains |= obj->read_domains; + else + obj_priv->dirty = 1; + + /* + * Flush the current write domain if + * the new read domains don't match. Invalidate + * any read domains which differ from the old + * write domain + */ + if (obj->write_domain && obj->write_domain != read_domains) { + flush_domains |= obj->write_domain; + invalidate_domains |= read_domains & ~obj->write_domain; + } + /* + * Invalidate any read caches which may have + * stale data. That is, any new read domains. + */ + invalidate_domains |= read_domains & ~obj->read_domains; + if ((flush_domains | invalidate_domains) & I915_GEM_DOMAIN_CPU) { +#if WATCH_BUF + DRM_INFO("%s: CPU domain flush %08x invalidate %08x\n", + __func__, flush_domains, invalidate_domains); +#endif + /* + * If we're invaliding the CPU cache and flushing a GPU cache, + * then pause for rendering so that the GPU caches will be + * flushed before the cpu cache is invalidated + */ + if ((invalidate_domains & I915_GEM_DOMAIN_CPU) && + (flush_domains & ~(I915_GEM_DOMAIN_CPU | + I915_GEM_DOMAIN_GTT))) { + ret = i915_gem_object_wait_rendering(obj); + if (ret) + return ret; + } + i915_gem_clflush_object(obj); + } + + if ((write_domain | flush_domains) != 0) + obj->write_domain = write_domain; + + /* If we're invalidating the CPU domain, clear the per-page CPU + * domain list as well. + */ + if (obj_priv->page_cpu_valid != NULL && + (write_domain != 0 || + read_domains & I915_GEM_DOMAIN_CPU)) { + drm_free(obj_priv->page_cpu_valid, obj->size / PAGE_SIZE, + DRM_MEM_DRIVER); + obj_priv->page_cpu_valid = NULL; + } + obj->read_domains = read_domains; + + dev->invalidate_domains |= invalidate_domains; + dev->flush_domains |= flush_domains; +#if WATCH_BUF + DRM_INFO("%s: read %08x write %08x invalidate %08x flush %08x\n", + __func__, + obj->read_domains, obj->write_domain, + dev->invalidate_domains, dev->flush_domains); +#endif + return 0; +} + +/** + * Set the read/write domain on a range of the object. + * + * Currently only implemented for CPU reads, otherwise drops to normal + * i915_gem_object_set_domain(). + */ +static int +i915_gem_object_set_domain_range(struct drm_gem_object *obj, + uint64_t offset, + uint64_t size, + uint32_t read_domains, + uint32_t write_domain) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int ret, i; + + if (obj->read_domains & I915_GEM_DOMAIN_CPU) + return 0; + + if (read_domains != I915_GEM_DOMAIN_CPU || + write_domain != 0) + return i915_gem_object_set_domain(obj, + read_domains, write_domain); + + /* Wait on any GPU rendering to the object to be flushed. */ + if (obj->write_domain & ~(I915_GEM_DOMAIN_CPU | I915_GEM_DOMAIN_GTT)) { + ret = i915_gem_object_wait_rendering(obj); + if (ret) + return ret; + } + + if (obj_priv->page_cpu_valid == NULL) { + obj_priv->page_cpu_valid = drm_calloc(1, obj->size / PAGE_SIZE, + DRM_MEM_DRIVER); + } + + /* Flush the cache on any pages that are still invalid from the CPU's + * perspective. + */ + for (i = offset / PAGE_SIZE; i <= (offset + size - 1) / PAGE_SIZE; i++) { + if (obj_priv->page_cpu_valid[i]) + continue; + + drm_clflush_pages(obj_priv->page_list + i, 1); + + obj_priv->page_cpu_valid[i] = 1; + } + + return 0; +} + +/** + * Once all of the objects have been set in the proper domain, + * perform the necessary flush and invalidate operations. + * + * Returns the write domains flushed, for use in flush tracking. + */ +static uint32_t +i915_gem_dev_set_domain(struct drm_device *dev) +{ + uint32_t flush_domains = dev->flush_domains; + + /* + * Now that all the buffers are synced to the proper domains, + * flush and invalidate the collected domains + */ + if (dev->invalidate_domains | dev->flush_domains) { +#if WATCH_EXEC + DRM_INFO("%s: invalidate_domains %08x flush_domains %08x\n", + __func__, + dev->invalidate_domains, + dev->flush_domains); +#endif + i915_gem_flush(dev, + dev->invalidate_domains, + dev->flush_domains); + dev->invalidate_domains = 0; + dev->flush_domains = 0; + } + + return flush_domains; +} + +/** + * Pin an object to the GTT and evaluate the relocations landing in it. + */ +static int +i915_gem_object_pin_and_relocate(struct drm_gem_object *obj, + struct drm_file *file_priv, + struct drm_i915_gem_exec_object *entry) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_relocation_entry reloc; + struct drm_i915_gem_relocation_entry __user *relocs; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int i, ret; + uint32_t last_reloc_offset = -1; + void *reloc_page = NULL; + + /* Choose the GTT offset for our buffer and put it there. */ + ret = i915_gem_object_pin(obj, (uint32_t) entry->alignment); + if (ret) + return ret; + + entry->offset = obj_priv->gtt_offset; + + relocs = (struct drm_i915_gem_relocation_entry __user *) + (uintptr_t) entry->relocs_ptr; + /* Apply the relocations, using the GTT aperture to avoid cache + * flushing requirements. + */ + for (i = 0; i < entry->relocation_count; i++) { + struct drm_gem_object *target_obj; + struct drm_i915_gem_object *target_obj_priv; + uint32_t reloc_val, reloc_offset, *reloc_entry; + int ret; + + ret = copy_from_user(&reloc, relocs + i, sizeof(reloc)); + if (ret != 0) { + i915_gem_object_unpin(obj); + return ret; + } + + target_obj = drm_gem_object_lookup(obj->dev, file_priv, + reloc.target_handle); + if (target_obj == NULL) { + i915_gem_object_unpin(obj); + return -EBADF; + } + target_obj_priv = target_obj->driver_private; + + /* The target buffer should have appeared before us in the + * exec_object list, so it should have a GTT space bound by now. + */ + if (target_obj_priv->gtt_space == NULL) { + DRM_ERROR("No GTT space found for object %d\n", + reloc.target_handle); + drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); + return -EINVAL; + } + + if (reloc.offset > obj->size - 4) { + DRM_ERROR("Relocation beyond object bounds: " + "obj %p target %d offset %d size %d.\n", + obj, reloc.target_handle, + (int) reloc.offset, (int) obj->size); + drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); + return -EINVAL; + } + if (reloc.offset & 3) { + DRM_ERROR("Relocation not 4-byte aligned: " + "obj %p target %d offset %d.\n", + obj, reloc.target_handle, + (int) reloc.offset); + drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); + return -EINVAL; + } + + if (reloc.write_domain && target_obj->pending_write_domain && + reloc.write_domain != target_obj->pending_write_domain) { + DRM_ERROR("Write domain conflict: " + "obj %p target %d offset %d " + "new %08x old %08x\n", + obj, reloc.target_handle, + (int) reloc.offset, + reloc.write_domain, + target_obj->pending_write_domain); + drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); + return -EINVAL; + } + +#if WATCH_RELOC + DRM_INFO("%s: obj %p offset %08x target %d " + "read %08x write %08x gtt %08x " + "presumed %08x delta %08x\n", + __func__, + obj, + (int) reloc.offset, + (int) reloc.target_handle, + (int) reloc.read_domains, + (int) reloc.write_domain, + (int) target_obj_priv->gtt_offset, + (int) reloc.presumed_offset, + reloc.delta); +#endif + + target_obj->pending_read_domains |= reloc.read_domains; + target_obj->pending_write_domain |= reloc.write_domain; + + /* If the relocation already has the right value in it, no + * more work needs to be done. + */ + if (target_obj_priv->gtt_offset == reloc.presumed_offset) { + drm_gem_object_unreference(target_obj); + continue; + } + + /* Now that we're going to actually write some data in, + * make sure that any rendering using this buffer's contents + * is completed. + */ + i915_gem_object_wait_rendering(obj); + + /* As we're writing through the gtt, flush + * any CPU writes before we write the relocations + */ + if (obj->write_domain & I915_GEM_DOMAIN_CPU) { + i915_gem_clflush_object(obj); + drm_agp_chipset_flush(dev); + obj->write_domain = 0; + } + + /* Map the page containing the relocation we're going to + * perform. + */ + reloc_offset = obj_priv->gtt_offset + reloc.offset; + if (reloc_page == NULL || + (last_reloc_offset & ~(PAGE_SIZE - 1)) != + (reloc_offset & ~(PAGE_SIZE - 1))) { + if (reloc_page != NULL) + iounmap(reloc_page); + + reloc_page = ioremap(dev->agp->base + + (reloc_offset & ~(PAGE_SIZE - 1)), + PAGE_SIZE); + last_reloc_offset = reloc_offset; + if (reloc_page == NULL) { + drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); + return -ENOMEM; + } + } + + reloc_entry = (uint32_t *)((char *)reloc_page + + (reloc_offset & (PAGE_SIZE - 1))); + reloc_val = target_obj_priv->gtt_offset + reloc.delta; + +#if WATCH_BUF + DRM_INFO("Applied relocation: %p@0x%08x %08x -> %08x\n", + obj, (unsigned int) reloc.offset, + readl(reloc_entry), reloc_val); +#endif + writel(reloc_val, reloc_entry); + + /* Write the updated presumed offset for this entry back out + * to the user. + */ + reloc.presumed_offset = target_obj_priv->gtt_offset; + ret = copy_to_user(relocs + i, &reloc, sizeof(reloc)); + if (ret != 0) { + drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); + return ret; + } + + drm_gem_object_unreference(target_obj); + } + + if (reloc_page != NULL) + iounmap(reloc_page); + +#if WATCH_BUF + if (0) + i915_gem_dump_object(obj, 128, __func__, ~0); +#endif + return 0; +} + +/** Dispatch a batchbuffer to the ring + */ +static int +i915_dispatch_gem_execbuffer(struct drm_device *dev, + struct drm_i915_gem_execbuffer *exec, + uint64_t exec_offset) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_clip_rect __user *boxes = (struct drm_clip_rect __user *) + (uintptr_t) exec->cliprects_ptr; + int nbox = exec->num_cliprects; + int i = 0, count; + uint32_t exec_start, exec_len; + RING_LOCALS; + + exec_start = (uint32_t) exec_offset + exec->batch_start_offset; + exec_len = (uint32_t) exec->batch_len; + + if ((exec_start | exec_len) & 0x7) { + DRM_ERROR("alignment\n"); + return -EINVAL; + } + + if (!exec_start) + return -EINVAL; + + count = nbox ? nbox : 1; + + for (i = 0; i < count; i++) { + if (i < nbox) { + int ret = i915_emit_box(dev, boxes, i, + exec->DR1, exec->DR4); + if (ret) + return ret; + } + + if (IS_I830(dev) || IS_845G(dev)) { + BEGIN_LP_RING(4); + OUT_RING(MI_BATCH_BUFFER); + OUT_RING(exec_start | MI_BATCH_NON_SECURE); + OUT_RING(exec_start + exec_len - 4); + OUT_RING(0); + ADVANCE_LP_RING(); + } else { + BEGIN_LP_RING(2); + if (IS_I965G(dev)) { + OUT_RING(MI_BATCH_BUFFER_START | + (2 << 6) | + MI_BATCH_NON_SECURE_I965); + OUT_RING(exec_start); + } else { + OUT_RING(MI_BATCH_BUFFER_START | + (2 << 6)); + OUT_RING(exec_start | MI_BATCH_NON_SECURE); + } + ADVANCE_LP_RING(); + } + } + + /* XXX breadcrumb */ + return 0; +} + +/* Throttle our rendering by waiting until the ring has completed our requests + * emitted over 20 msec ago. + * + * This should get us reasonable parallelism between CPU and GPU but also + * relatively low latency when blocking on a particular request to finish. + */ +static int +i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file_priv) +{ + struct drm_i915_file_private *i915_file_priv = file_priv->driver_priv; + int ret = 0; + uint32_t seqno; + + mutex_lock(&dev->struct_mutex); + seqno = i915_file_priv->mm.last_gem_throttle_seqno; + i915_file_priv->mm.last_gem_throttle_seqno = + i915_file_priv->mm.last_gem_seqno; + if (seqno) + ret = i915_wait_request(dev, seqno); + mutex_unlock(&dev->struct_mutex); + return ret; +} + +int +i915_gem_execbuffer(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_file_private *i915_file_priv = file_priv->driver_priv; + struct drm_i915_gem_execbuffer *args = data; + struct drm_i915_gem_exec_object *exec_list = NULL; + struct drm_gem_object **object_list = NULL; + struct drm_gem_object *batch_obj; + int ret, i, pinned = 0; + uint64_t exec_offset; + uint32_t seqno, flush_domains; + +#if WATCH_EXEC + DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n", + (int) args->buffers_ptr, args->buffer_count, args->batch_len); +#endif + + /* Copy in the exec list from userland */ + exec_list = drm_calloc(sizeof(*exec_list), args->buffer_count, + DRM_MEM_DRIVER); + object_list = drm_calloc(sizeof(*object_list), args->buffer_count, + DRM_MEM_DRIVER); + if (exec_list == NULL || object_list == NULL) { + DRM_ERROR("Failed to allocate exec or object list " + "for %d buffers\n", + args->buffer_count); + ret = -ENOMEM; + goto pre_mutex_err; + } + ret = copy_from_user(exec_list, + (struct drm_i915_relocation_entry __user *) + (uintptr_t) args->buffers_ptr, + sizeof(*exec_list) * args->buffer_count); + if (ret != 0) { + DRM_ERROR("copy %d exec entries failed %d\n", + args->buffer_count, ret); + goto pre_mutex_err; + } + + mutex_lock(&dev->struct_mutex); + + i915_verify_inactive(dev, __FILE__, __LINE__); + + if (dev_priv->mm.wedged) { + DRM_ERROR("Execbuf while wedged\n"); + mutex_unlock(&dev->struct_mutex); + return -EIO; + } + + if (dev_priv->mm.suspended) { + DRM_ERROR("Execbuf while VT-switched.\n"); + mutex_unlock(&dev->struct_mutex); + return -EBUSY; + } + + /* Zero the gloabl flush/invalidate flags. These + * will be modified as each object is bound to the + * gtt + */ + dev->invalidate_domains = 0; + dev->flush_domains = 0; + + /* Look up object handles and perform the relocations */ + for (i = 0; i < args->buffer_count; i++) { + object_list[i] = drm_gem_object_lookup(dev, file_priv, + exec_list[i].handle); + if (object_list[i] == NULL) { + DRM_ERROR("Invalid object handle %d at index %d\n", + exec_list[i].handle, i); + ret = -EBADF; + goto err; + } + + object_list[i]->pending_read_domains = 0; + object_list[i]->pending_write_domain = 0; + ret = i915_gem_object_pin_and_relocate(object_list[i], + file_priv, + &exec_list[i]); + if (ret) { + DRM_ERROR("object bind and relocate failed %d\n", ret); + goto err; + } + pinned = i + 1; + } + + /* Set the pending read domains for the batch buffer to COMMAND */ + batch_obj = object_list[args->buffer_count-1]; + batch_obj->pending_read_domains = I915_GEM_DOMAIN_COMMAND; + batch_obj->pending_write_domain = 0; + + i915_verify_inactive(dev, __FILE__, __LINE__); + + for (i = 0; i < args->buffer_count; i++) { + struct drm_gem_object *obj = object_list[i]; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + if (obj_priv->gtt_space == NULL) { + /* We evicted the buffer in the process of validating + * our set of buffers in. We could try to recover by + * kicking them everything out and trying again from + * the start. + */ + ret = -ENOMEM; + goto err; + } + + /* make sure all previous memory operations have passed */ + ret = i915_gem_object_set_domain(obj, + obj->pending_read_domains, + obj->pending_write_domain); + if (ret) + goto err; + } + + i915_verify_inactive(dev, __FILE__, __LINE__); + + /* Flush/invalidate caches and chipset buffer */ + flush_domains = i915_gem_dev_set_domain(dev); + + i915_verify_inactive(dev, __FILE__, __LINE__); + +#if WATCH_COHERENCY + for (i = 0; i < args->buffer_count; i++) { + i915_gem_object_check_coherency(object_list[i], + exec_list[i].handle); + } +#endif + + exec_offset = exec_list[args->buffer_count - 1].offset; + +#if WATCH_EXEC + i915_gem_dump_object(object_list[args->buffer_count - 1], + args->batch_len, + __func__, + ~0); +#endif + + (void)i915_add_request(dev, flush_domains); + + /* Exec the batchbuffer */ + ret = i915_dispatch_gem_execbuffer(dev, args, exec_offset); + if (ret) { + DRM_ERROR("dispatch failed %d\n", ret); + goto err; + } + + /* + * Ensure that the commands in the batch buffer are + * finished before the interrupt fires + */ + flush_domains = i915_retire_commands(dev); + + i915_verify_inactive(dev, __FILE__, __LINE__); + + /* + * Get a seqno representing the execution of the current buffer, + * which we can wait on. We would like to mitigate these interrupts, + * likely by only creating seqnos occasionally (so that we have + * *some* interrupts representing completion of buffers that we can + * wait on when trying to clear up gtt space). + */ + seqno = i915_add_request(dev, flush_domains); + BUG_ON(seqno == 0); + i915_file_priv->mm.last_gem_seqno = seqno; + for (i = 0; i < args->buffer_count; i++) { + struct drm_gem_object *obj = object_list[i]; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + i915_gem_object_move_to_active(obj); + obj_priv->last_rendering_seqno = seqno; +#if WATCH_LRU + DRM_INFO("%s: move to exec list %p\n", __func__, obj); +#endif + } +#if WATCH_LRU + i915_dump_lru(dev, __func__); +#endif + + i915_verify_inactive(dev, __FILE__, __LINE__); + + /* Copy the new buffer offsets back to the user's exec list. */ + ret = copy_to_user((struct drm_i915_relocation_entry __user *) + (uintptr_t) args->buffers_ptr, + exec_list, + sizeof(*exec_list) * args->buffer_count); + if (ret) + DRM_ERROR("failed to copy %d exec entries " + "back to user (%d)\n", + args->buffer_count, ret); +err: + if (object_list != NULL) { + for (i = 0; i < pinned; i++) + i915_gem_object_unpin(object_list[i]); + + for (i = 0; i < args->buffer_count; i++) + drm_gem_object_unreference(object_list[i]); + } + mutex_unlock(&dev->struct_mutex); + +pre_mutex_err: + drm_free(object_list, sizeof(*object_list) * args->buffer_count, + DRM_MEM_DRIVER); + drm_free(exec_list, sizeof(*exec_list) * args->buffer_count, + DRM_MEM_DRIVER); + + return ret; +} + +int +i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int ret; + + i915_verify_inactive(dev, __FILE__, __LINE__); + if (obj_priv->gtt_space == NULL) { + ret = i915_gem_object_bind_to_gtt(obj, alignment); + if (ret != 0) { + DRM_ERROR("Failure to bind: %d", ret); + return ret; + } + } + obj_priv->pin_count++; + + /* If the object is not active and not pending a flush, + * remove it from the inactive list + */ + if (obj_priv->pin_count == 1) { + atomic_inc(&dev->pin_count); + atomic_add(obj->size, &dev->pin_memory); + if (!obj_priv->active && + (obj->write_domain & ~(I915_GEM_DOMAIN_CPU | + I915_GEM_DOMAIN_GTT)) == 0 && + !list_empty(&obj_priv->list)) + list_del_init(&obj_priv->list); + } + i915_verify_inactive(dev, __FILE__, __LINE__); + + return 0; +} + +void +i915_gem_object_unpin(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + i915_verify_inactive(dev, __FILE__, __LINE__); + obj_priv->pin_count--; + BUG_ON(obj_priv->pin_count < 0); + BUG_ON(obj_priv->gtt_space == NULL); + + /* If the object is no longer pinned, and is + * neither active nor being flushed, then stick it on + * the inactive list + */ + if (obj_priv->pin_count == 0) { + if (!obj_priv->active && + (obj->write_domain & ~(I915_GEM_DOMAIN_CPU | + I915_GEM_DOMAIN_GTT)) == 0) + list_move_tail(&obj_priv->list, + &dev_priv->mm.inactive_list); + atomic_dec(&dev->pin_count); + atomic_sub(obj->size, &dev->pin_memory); + } + i915_verify_inactive(dev, __FILE__, __LINE__); +} + +int +i915_gem_pin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_pin *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret; + + mutex_lock(&dev->struct_mutex); + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + DRM_ERROR("Bad handle in i915_gem_pin_ioctl(): %d\n", + args->handle); + mutex_unlock(&dev->struct_mutex); + return -EBADF; + } + obj_priv = obj->driver_private; + + ret = i915_gem_object_pin(obj, args->alignment); + if (ret != 0) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + + /* XXX - flush the CPU caches for pinned objects + * as the X server doesn't manage domains yet + */ + if (obj->write_domain & I915_GEM_DOMAIN_CPU) { + i915_gem_clflush_object(obj); + drm_agp_chipset_flush(dev); + obj->write_domain = 0; + } + args->offset = obj_priv->gtt_offset; + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +int +i915_gem_unpin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_pin *args = data; + struct drm_gem_object *obj; + + mutex_lock(&dev->struct_mutex); + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + DRM_ERROR("Bad handle in i915_gem_unpin_ioctl(): %d\n", + args->handle); + mutex_unlock(&dev->struct_mutex); + return -EBADF; + } + + i915_gem_object_unpin(obj); + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return 0; +} + +int +i915_gem_busy_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_busy *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + mutex_lock(&dev->struct_mutex); + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + DRM_ERROR("Bad handle in i915_gem_busy_ioctl(): %d\n", + args->handle); + mutex_unlock(&dev->struct_mutex); + return -EBADF; + } + + obj_priv = obj->driver_private; + args->busy = obj_priv->active; + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return 0; +} + +int +i915_gem_throttle_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + return i915_gem_ring_throttle(dev, file_priv); +} + +int i915_gem_init_object(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv; + + obj_priv = drm_calloc(1, sizeof(*obj_priv), DRM_MEM_DRIVER); + if (obj_priv == NULL) + return -ENOMEM; + + /* + * We've just allocated pages from the kernel, + * so they've just been written by the CPU with + * zeros. They'll need to be clflushed before we + * use them with the GPU. + */ + obj->write_domain = I915_GEM_DOMAIN_CPU; + obj->read_domains = I915_GEM_DOMAIN_CPU; + + obj->driver_private = obj_priv; + obj_priv->obj = obj; + INIT_LIST_HEAD(&obj_priv->list); + return 0; +} + +void i915_gem_free_object(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + while (obj_priv->pin_count > 0) + i915_gem_object_unpin(obj); + + i915_gem_object_unbind(obj); + + drm_free(obj_priv->page_cpu_valid, 1, DRM_MEM_DRIVER); + drm_free(obj->driver_private, 1, DRM_MEM_DRIVER); +} + +static int +i915_gem_set_domain(struct drm_gem_object *obj, + struct drm_file *file_priv, + uint32_t read_domains, + uint32_t write_domain) +{ + struct drm_device *dev = obj->dev; + int ret; + uint32_t flush_domains; + + BUG_ON(!mutex_is_locked(&dev->struct_mutex)); + + ret = i915_gem_object_set_domain(obj, read_domains, write_domain); + if (ret) + return ret; + flush_domains = i915_gem_dev_set_domain(obj->dev); + + if (flush_domains & ~(I915_GEM_DOMAIN_CPU|I915_GEM_DOMAIN_GTT)) + (void) i915_add_request(dev, flush_domains); + + return 0; +} + +/** Unbinds all objects that are on the given buffer list. */ +static int +i915_gem_evict_from_list(struct drm_device *dev, struct list_head *head) +{ + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret; + + while (!list_empty(head)) { + obj_priv = list_first_entry(head, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + + if (obj_priv->pin_count != 0) { + DRM_ERROR("Pinned object in unbind list\n"); + mutex_unlock(&dev->struct_mutex); + return -EINVAL; + } + + ret = i915_gem_object_unbind(obj); + if (ret != 0) { + DRM_ERROR("Error unbinding object in LeaveVT: %d\n", + ret); + mutex_unlock(&dev->struct_mutex); + return ret; + } + } + + + return 0; +} + +static int +i915_gem_idle(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t seqno, cur_seqno, last_seqno; + int stuck, ret; + + if (dev_priv->mm.suspended) + return 0; + + /* Hack! Don't let anybody do execbuf while we don't control the chip. + * We need to replace this with a semaphore, or something. + */ + dev_priv->mm.suspended = 1; + + i915_kernel_lost_context(dev); + + /* Flush the GPU along with all non-CPU write domains + */ + i915_gem_flush(dev, ~(I915_GEM_DOMAIN_CPU|I915_GEM_DOMAIN_GTT), + ~(I915_GEM_DOMAIN_CPU|I915_GEM_DOMAIN_GTT)); + seqno = i915_add_request(dev, ~(I915_GEM_DOMAIN_CPU | + I915_GEM_DOMAIN_GTT)); + + if (seqno == 0) { + mutex_unlock(&dev->struct_mutex); + return -ENOMEM; + } + + dev_priv->mm.waiting_gem_seqno = seqno; + last_seqno = 0; + stuck = 0; + for (;;) { + cur_seqno = i915_get_gem_seqno(dev); + if (i915_seqno_passed(cur_seqno, seqno)) + break; + if (last_seqno == cur_seqno) { + if (stuck++ > 100) { + DRM_ERROR("hardware wedged\n"); + dev_priv->mm.wedged = 1; + DRM_WAKEUP(&dev_priv->irq_queue); + break; + } + } + msleep(10); + last_seqno = cur_seqno; + } + dev_priv->mm.waiting_gem_seqno = 0; + + i915_gem_retire_requests(dev); + + /* Active and flushing should now be empty as we've + * waited for a sequence higher than any pending execbuffer + */ + BUG_ON(!list_empty(&dev_priv->mm.active_list)); + BUG_ON(!list_empty(&dev_priv->mm.flushing_list)); + + /* Request should now be empty as we've also waited + * for the last request in the list + */ + BUG_ON(!list_empty(&dev_priv->mm.request_list)); + + /* Move all buffers out of the GTT. */ + ret = i915_gem_evict_from_list(dev, &dev_priv->mm.inactive_list); + if (ret) + return ret; + + BUG_ON(!list_empty(&dev_priv->mm.active_list)); + BUG_ON(!list_empty(&dev_priv->mm.flushing_list)); + BUG_ON(!list_empty(&dev_priv->mm.inactive_list)); + BUG_ON(!list_empty(&dev_priv->mm.request_list)); + return 0; +} + +static int +i915_gem_init_hws(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret; + + /* If we need a physical address for the status page, it's already + * initialized at driver load time. + */ + if (!I915_NEED_GFX_HWS(dev)) + return 0; + + obj = drm_gem_object_alloc(dev, 4096); + if (obj == NULL) { + DRM_ERROR("Failed to allocate status page\n"); + return -ENOMEM; + } + obj_priv = obj->driver_private; + + ret = i915_gem_object_pin(obj, 4096); + if (ret != 0) { + drm_gem_object_unreference(obj); + return ret; + } + + dev_priv->status_gfx_addr = obj_priv->gtt_offset; + dev_priv->hws_map.offset = dev->agp->base + obj_priv->gtt_offset; + dev_priv->hws_map.size = 4096; + dev_priv->hws_map.type = 0; + dev_priv->hws_map.flags = 0; + dev_priv->hws_map.mtrr = 0; + + drm_core_ioremap(&dev_priv->hws_map, dev); + if (dev_priv->hws_map.handle == NULL) { + DRM_ERROR("Failed to map status page.\n"); + memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map)); + drm_gem_object_unreference(obj); + return -EINVAL; + } + dev_priv->hws_obj = obj; + dev_priv->hw_status_page = dev_priv->hws_map.handle; + memset(dev_priv->hw_status_page, 0, PAGE_SIZE); + I915_WRITE(HWS_PGA, dev_priv->status_gfx_addr); + DRM_DEBUG("hws offset: 0x%08x\n", dev_priv->status_gfx_addr); + + return 0; +} + +static int +i915_gem_init_ringbuffer(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret; + + ret = i915_gem_init_hws(dev); + if (ret != 0) + return ret; + + obj = drm_gem_object_alloc(dev, 128 * 1024); + if (obj == NULL) { + DRM_ERROR("Failed to allocate ringbuffer\n"); + return -ENOMEM; + } + obj_priv = obj->driver_private; + + ret = i915_gem_object_pin(obj, 4096); + if (ret != 0) { + drm_gem_object_unreference(obj); + return ret; + } + + /* Set up the kernel mapping for the ring. */ + dev_priv->ring.Size = obj->size; + dev_priv->ring.tail_mask = obj->size - 1; + + dev_priv->ring.map.offset = dev->agp->base + obj_priv->gtt_offset; + dev_priv->ring.map.size = obj->size; + dev_priv->ring.map.type = 0; + dev_priv->ring.map.flags = 0; + dev_priv->ring.map.mtrr = 0; + + drm_core_ioremap(&dev_priv->ring.map, dev); + if (dev_priv->ring.map.handle == NULL) { + DRM_ERROR("Failed to map ringbuffer.\n"); + memset(&dev_priv->ring, 0, sizeof(dev_priv->ring)); + drm_gem_object_unreference(obj); + return -EINVAL; + } + dev_priv->ring.ring_obj = obj; + dev_priv->ring.virtual_start = dev_priv->ring.map.handle; + + /* Stop the ring if it's running. */ + I915_WRITE(PRB0_CTL, 0); + I915_WRITE(PRB0_HEAD, 0); + I915_WRITE(PRB0_TAIL, 0); + I915_WRITE(PRB0_START, 0); + + /* Initialize the ring. */ + I915_WRITE(PRB0_START, obj_priv->gtt_offset); + I915_WRITE(PRB0_CTL, + ((obj->size - 4096) & RING_NR_PAGES) | + RING_NO_REPORT | + RING_VALID); + + /* Update our cache of the ring state */ + i915_kernel_lost_context(dev); + + return 0; +} + +static void +i915_gem_cleanup_ringbuffer(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + + if (dev_priv->ring.ring_obj == NULL) + return; + + drm_core_ioremapfree(&dev_priv->ring.map, dev); + + i915_gem_object_unpin(dev_priv->ring.ring_obj); + drm_gem_object_unreference(dev_priv->ring.ring_obj); + dev_priv->ring.ring_obj = NULL; + memset(&dev_priv->ring, 0, sizeof(dev_priv->ring)); + + if (dev_priv->hws_obj != NULL) { + i915_gem_object_unpin(dev_priv->hws_obj); + drm_gem_object_unreference(dev_priv->hws_obj); + dev_priv->hws_obj = NULL; + memset(&dev_priv->hws_map, 0, sizeof(dev_priv->hws_map)); + + /* Write high address into HWS_PGA when disabling. */ + I915_WRITE(HWS_PGA, 0x1ffff000); + } +} + +int +i915_gem_entervt_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + int ret; + + if (dev_priv->mm.wedged) { + DRM_ERROR("Reenabling wedged hardware, good luck\n"); + dev_priv->mm.wedged = 0; + } + + ret = i915_gem_init_ringbuffer(dev); + if (ret != 0) + return ret; + + mutex_lock(&dev->struct_mutex); + BUG_ON(!list_empty(&dev_priv->mm.active_list)); + BUG_ON(!list_empty(&dev_priv->mm.flushing_list)); + BUG_ON(!list_empty(&dev_priv->mm.inactive_list)); + BUG_ON(!list_empty(&dev_priv->mm.request_list)); + dev_priv->mm.suspended = 0; + mutex_unlock(&dev->struct_mutex); + return 0; +} + +int +i915_gem_leavevt_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + int ret; + + mutex_lock(&dev->struct_mutex); + ret = i915_gem_idle(dev); + if (ret == 0) + i915_gem_cleanup_ringbuffer(dev); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +void +i915_gem_lastclose(struct drm_device *dev) +{ + int ret; + drm_i915_private_t *dev_priv = dev->dev_private; + + mutex_lock(&dev->struct_mutex); + + if (dev_priv->ring.ring_obj != NULL) { + ret = i915_gem_idle(dev); + if (ret) + DRM_ERROR("failed to idle hardware: %d\n", ret); + + i915_gem_cleanup_ringbuffer(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); +} diff --git a/drivers/gpu/drm/i915/i915_gem_debug.c b/drivers/gpu/drm/i915/i915_gem_debug.c new file mode 100644 index 000000000000..131c088f8c8a --- /dev/null +++ b/drivers/gpu/drm/i915/i915_gem_debug.c @@ -0,0 +1,201 @@ +/* + * 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: + * Keith Packard <keithp@keithp.com> + * + */ + +#include "drmP.h" +#include "drm.h" +#include "i915_drm.h" +#include "i915_drv.h" + +#if WATCH_INACTIVE +void +i915_verify_inactive(struct drm_device *dev, char *file, int line) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) { + obj = obj_priv->obj; + if (obj_priv->pin_count || obj_priv->active || + (obj->write_domain & ~(I915_GEM_DOMAIN_CPU | + I915_GEM_DOMAIN_GTT))) + DRM_ERROR("inactive %p (p %d a %d w %x) %s:%d\n", + obj, + obj_priv->pin_count, obj_priv->active, + obj->write_domain, file, line); + } +} +#endif /* WATCH_INACTIVE */ + + +#if WATCH_BUF | WATCH_EXEC | WATCH_PWRITE +static void +i915_gem_dump_page(struct page *page, uint32_t start, uint32_t end, + uint32_t bias, uint32_t mark) +{ + uint32_t *mem = kmap_atomic(page, KM_USER0); + int i; + for (i = start; i < end; i += 4) + DRM_INFO("%08x: %08x%s\n", + (int) (bias + i), mem[i / 4], + (bias + i == mark) ? " ********" : ""); + kunmap_atomic(mem, KM_USER0); + /* give syslog time to catch up */ + msleep(1); +} + +void +i915_gem_dump_object(struct drm_gem_object *obj, int len, + const char *where, uint32_t mark) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page; + + DRM_INFO("%s: object at offset %08x\n", where, obj_priv->gtt_offset); + for (page = 0; page < (len + PAGE_SIZE-1) / PAGE_SIZE; page++) { + int page_len, chunk, chunk_len; + + page_len = len - page * PAGE_SIZE; + if (page_len > PAGE_SIZE) + page_len = PAGE_SIZE; + + for (chunk = 0; chunk < page_len; chunk += 128) { + chunk_len = page_len - chunk; + if (chunk_len > 128) + chunk_len = 128; + i915_gem_dump_page(obj_priv->page_list[page], + chunk, chunk + chunk_len, + obj_priv->gtt_offset + + page * PAGE_SIZE, + mark); + } + } +} +#endif + +#if WATCH_LRU +void +i915_dump_lru(struct drm_device *dev, const char *where) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv; + + DRM_INFO("active list %s {\n", where); + list_for_each_entry(obj_priv, &dev_priv->mm.active_list, + list) + { + DRM_INFO(" %p: %08x\n", obj_priv, + obj_priv->last_rendering_seqno); + } + DRM_INFO("}\n"); + DRM_INFO("flushing list %s {\n", where); + list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, + list) + { + DRM_INFO(" %p: %08x\n", obj_priv, + obj_priv->last_rendering_seqno); + } + DRM_INFO("}\n"); + DRM_INFO("inactive %s {\n", where); + list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) { + DRM_INFO(" %p: %08x\n", obj_priv, + obj_priv->last_rendering_seqno); + } + DRM_INFO("}\n"); +} +#endif + + +#if WATCH_COHERENCY +void +i915_gem_object_check_coherency(struct drm_gem_object *obj, int handle) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page; + uint32_t *gtt_mapping; + uint32_t *backing_map = NULL; + int bad_count = 0; + + DRM_INFO("%s: checking coherency of object %p@0x%08x (%d, %dkb):\n", + __func__, obj, obj_priv->gtt_offset, handle, + obj->size / 1024); + + gtt_mapping = ioremap(dev->agp->base + obj_priv->gtt_offset, + obj->size); + if (gtt_mapping == NULL) { + DRM_ERROR("failed to map GTT space\n"); + return; + } + + for (page = 0; page < obj->size / PAGE_SIZE; page++) { + int i; + + backing_map = kmap_atomic(obj_priv->page_list[page], KM_USER0); + + if (backing_map == NULL) { + DRM_ERROR("failed to map backing page\n"); + goto out; + } + + for (i = 0; i < PAGE_SIZE / 4; i++) { + uint32_t cpuval = backing_map[i]; + uint32_t gttval = readl(gtt_mapping + + page * 1024 + i); + + if (cpuval != gttval) { + DRM_INFO("incoherent CPU vs GPU at 0x%08x: " + "0x%08x vs 0x%08x\n", + (int)(obj_priv->gtt_offset + + page * PAGE_SIZE + i * 4), + cpuval, gttval); + if (bad_count++ >= 8) { + DRM_INFO("...\n"); + goto out; + } + } + } + kunmap_atomic(backing_map, KM_USER0); + backing_map = NULL; + } + + out: + if (backing_map != NULL) + kunmap_atomic(backing_map, KM_USER0); + iounmap(gtt_mapping); + + /* give syslog time to catch up */ + msleep(1); + + /* Directly flush the object, since we just loaded values with the CPU + * from the backing pages and we don't want to disturb the cache + * management that we're trying to observe. + */ + + i915_gem_clflush_object(obj); +} +#endif diff --git a/drivers/gpu/drm/i915/i915_gem_proc.c b/drivers/gpu/drm/i915/i915_gem_proc.c new file mode 100644 index 000000000000..15d4160415b0 --- /dev/null +++ b/drivers/gpu/drm/i915/i915_gem_proc.c @@ -0,0 +1,292 @@ +/* + * 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> + * Keith Packard <keithp@keithp.com> + * + */ + +#include "drmP.h" +#include "drm.h" +#include "i915_drm.h" +#include "i915_drv.h" + +static int i915_gem_active_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("Active:\n"); + list_for_each_entry(obj_priv, &dev_priv->mm.active_list, + list) + { + struct drm_gem_object *obj = obj_priv->obj; + if (obj->name) { + DRM_PROC_PRINT(" %p(%d): %08x %08x %d\n", + obj, obj->name, + obj->read_domains, obj->write_domain, + obj_priv->last_rendering_seqno); + } else { + DRM_PROC_PRINT(" %p: %08x %08x %d\n", + obj, + obj->read_domains, obj->write_domain, + obj_priv->last_rendering_seqno); + } + } + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + +static int i915_gem_flushing_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("Flushing:\n"); + list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, + list) + { + struct drm_gem_object *obj = obj_priv->obj; + if (obj->name) { + DRM_PROC_PRINT(" %p(%d): %08x %08x %d\n", + obj, obj->name, + obj->read_domains, obj->write_domain, + obj_priv->last_rendering_seqno); + } else { + DRM_PROC_PRINT(" %p: %08x %08x %d\n", obj, + obj->read_domains, obj->write_domain, + obj_priv->last_rendering_seqno); + } + } + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + +static int i915_gem_inactive_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("Inactive:\n"); + list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, + list) + { + struct drm_gem_object *obj = obj_priv->obj; + if (obj->name) { + DRM_PROC_PRINT(" %p(%d): %08x %08x %d\n", + obj, obj->name, + obj->read_domains, obj->write_domain, + obj_priv->last_rendering_seqno); + } else { + DRM_PROC_PRINT(" %p: %08x %08x %d\n", obj, + obj->read_domains, obj->write_domain, + obj_priv->last_rendering_seqno); + } + } + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + +static int i915_gem_request_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_request *gem_request; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("Request:\n"); + list_for_each_entry(gem_request, &dev_priv->mm.request_list, + list) + { + DRM_PROC_PRINT(" %d @ %d %08x\n", + gem_request->seqno, + (int) (jiffies - gem_request->emitted_jiffies), + gem_request->flush_domains); + } + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + +static int i915_gem_seqno_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("Current sequence: %d\n", i915_get_gem_seqno(dev)); + DRM_PROC_PRINT("Waiter sequence: %d\n", + dev_priv->mm.waiting_gem_seqno); + DRM_PROC_PRINT("IRQ sequence: %d\n", dev_priv->mm.irq_gem_seqno); + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + + +static int i915_interrupt_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT("Interrupt enable: %08x\n", + I915_READ(IER)); + DRM_PROC_PRINT("Interrupt identity: %08x\n", + I915_READ(IIR)); + DRM_PROC_PRINT("Interrupt mask: %08x\n", + I915_READ(IMR)); + DRM_PROC_PRINT("Pipe A stat: %08x\n", + I915_READ(PIPEASTAT)); + DRM_PROC_PRINT("Pipe B stat: %08x\n", + I915_READ(PIPEBSTAT)); + DRM_PROC_PRINT("Interrupts received: %d\n", + atomic_read(&dev_priv->irq_received)); + DRM_PROC_PRINT("Current sequence: %d\n", + i915_get_gem_seqno(dev)); + DRM_PROC_PRINT("Waiter sequence: %d\n", + dev_priv->mm.waiting_gem_seqno); + DRM_PROC_PRINT("IRQ sequence: %d\n", + dev_priv->mm.irq_gem_seqno); + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + +static struct drm_proc_list { + /** file name */ + const char *name; + /** proc callback*/ + int (*f) (char *, char **, off_t, int, int *, void *); +} i915_gem_proc_list[] = { + {"i915_gem_active", i915_gem_active_info}, + {"i915_gem_flushing", i915_gem_flushing_info}, + {"i915_gem_inactive", i915_gem_inactive_info}, + {"i915_gem_request", i915_gem_request_info}, + {"i915_gem_seqno", i915_gem_seqno_info}, + {"i915_gem_interrupt", i915_interrupt_info}, +}; + +#define I915_GEM_PROC_ENTRIES ARRAY_SIZE(i915_gem_proc_list) + +int i915_gem_proc_init(struct drm_minor *minor) +{ + struct proc_dir_entry *ent; + int i, j; + + for (i = 0; i < I915_GEM_PROC_ENTRIES; i++) { + ent = create_proc_entry(i915_gem_proc_list[i].name, + S_IFREG | S_IRUGO, minor->dev_root); + if (!ent) { + DRM_ERROR("Cannot create /proc/dri/.../%s\n", + i915_gem_proc_list[i].name); + for (j = 0; j < i; j++) + remove_proc_entry(i915_gem_proc_list[i].name, + minor->dev_root); + return -1; + } + ent->read_proc = i915_gem_proc_list[i].f; + ent->data = minor; + } + return 0; +} + +void i915_gem_proc_cleanup(struct drm_minor *minor) +{ + int i; + + if (!minor->dev_root) + return; + + for (i = 0; i < I915_GEM_PROC_ENTRIES; i++) + remove_proc_entry(i915_gem_proc_list[i].name, minor->dev_root); +} diff --git a/drivers/gpu/drm/i915/i915_gem_tiling.c b/drivers/gpu/drm/i915/i915_gem_tiling.c new file mode 100644 index 000000000000..0c1b3a0834e1 --- /dev/null +++ b/drivers/gpu/drm/i915/i915_gem_tiling.c @@ -0,0 +1,256 @@ +/* + * 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; + uint32_t swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN; + uint32_t swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN; + + if (!IS_I9XX(dev)) { + /* As far as we know, the 865 doesn't have these bit 6 + * swizzling issues. + */ + swizzle_x = I915_BIT_6_SWIZZLE_NONE; + swizzle_y = I915_BIT_6_SWIZZLE_NONE; + } else if (!IS_I965G(dev) || IS_I965GM(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 = I915_READ(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; + } + if (dcc == 0xffffffff) { + DRM_ERROR("Couldn't read from MCHBAR. " + "Disabling tiling.\n"); + swizzle_x = I915_BIT_6_SWIZZLE_UNKNOWN; + swizzle_y = I915_BIT_6_SWIZZLE_UNKNOWN; + } + } else { + /* The 965, G33, and newer, have a very flexible memory + * configuration. It will enable dual-channel mode + * (interleaving) on as much memory as it can, and the GPU + * will additionally sometimes enable different bit 6 + * swizzling for tiled objects from the CPU. + * + * Here's what I found on the G965: + * slot fill memory size swizzling + * 0A 0B 1A 1B 1-ch 2-ch + * 512 0 0 0 512 0 O + * 512 0 512 0 16 1008 X + * 512 0 0 512 16 1008 X + * 0 512 0 512 16 1008 X + * 1024 1024 1024 0 2048 1024 O + * + * We could probably detect this based on either the DRB + * matching, which was the case for the swizzling required in + * the table above, or from the 1-ch value being less than + * the minimum size of a rank. + */ + if (I915_READ16(C0DRB3) != I915_READ16(C1DRB3)) { + 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; + } + } + + 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; +} diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c index f8759597233b..f295bdf16e2d 100644 --- a/drivers/gpu/drm/i915/i915_irq.c +++ b/drivers/gpu/drm/i915/i915_irq.c @@ -407,15 +407,20 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) I915_WRITE(PIPEBSTAT, pipeb_stats); } - if (iir & I915_ASLE_INTERRUPT) - opregion_asle_intr(dev); + I915_WRITE(IIR, iir); + if (dev->pdev->msi_enabled) + I915_WRITE(IMR, dev_priv->irq_mask_reg); + (void) I915_READ(IIR); /* Flush posted writes */ dev_priv->sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv); - if (dev->pdev->msi_enabled) - I915_WRITE(IMR, dev_priv->irq_mask_reg); - I915_WRITE(IIR, iir); - (void) I915_READ(IIR); + if (iir & I915_USER_INTERRUPT) { + dev_priv->mm.irq_gem_seqno = i915_get_gem_seqno(dev); + DRM_WAKEUP(&dev_priv->irq_queue); + } + + if (iir & I915_ASLE_INTERRUPT) + opregion_asle_intr(dev); if (vblank && dev_priv->swaps_pending > 0) drm_locked_tasklet(dev, i915_vblank_tasklet); @@ -449,7 +454,7 @@ static int i915_emit_irq(struct drm_device * dev) return dev_priv->counter; } -static void i915_user_irq_get(struct drm_device *dev) +void i915_user_irq_get(struct drm_device *dev) { drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h index 43ad2cb8b9f5..5c2d9f206d05 100644 --- a/drivers/gpu/drm/i915/i915_reg.h +++ b/drivers/gpu/drm/i915/i915_reg.h @@ -25,19 +25,6 @@ #ifndef _I915_REG_H_ #define _I915_REG_H_ -/* 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) - /* * The Bridge device's PCI config space has information about the * fb aperture size and the amount of pre-reserved memory. @@ -516,6 +503,30 @@ #define PALETTE_A 0x0a000 #define PALETTE_B 0x0a800 +/* MCH MMIO space */ + +/* + * MCHBAR mirror. + * + * This mirrors the MCHBAR MMIO space whose location is determined by + * device 0 function 0's pci config register 0x44 or 0x48 and matches it in + * every way. It is not accessible from the CP register read instructions. + * + */ +#define MCHBAR_MIRROR_BASE 0x10000 + +/** 915-945 and GM965 MCH register controlling DRAM channel access */ +#define DCC 0x10200 +#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 C0DRB3 0x10206 +#define C1DRB3 0x10606 + /* * Overlay regs */ |