From dac3bcb414a21a77847c96740a1578f3488c774f Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 11 Jun 2008 11:28:20 -0700 Subject: [gem] Remove carefully-sprinkled i915_kernel_lost_context(). They are not unnecessary since the kernel's the only thing touching the ring. --- linux-core/i915_gem.c | 5 ----- 1 file changed, 5 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 2564f418..3d47ec1a 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1404,7 +1404,6 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n", (int) args->buffers_ptr, args->buffer_count, args->batch_len); #endif - i915_kernel_lost_context(dev); /* Copy in the exec list from userland */ exec_list = drm_calloc(sizeof(*exec_list), args->buffer_count, @@ -1614,7 +1613,6 @@ i915_gem_pin_ioctl(struct drm_device *dev, void *data, mutex_lock(&dev->struct_mutex); - i915_kernel_lost_context(dev); obj = drm_gem_object_lookup(dev, file_priv, args->handle); if (obj == NULL) { DRM_ERROR("Bad handle in i915_gem_pin_ioctl(): %d\n", @@ -1647,7 +1645,6 @@ i915_gem_unpin_ioctl(struct drm_device *dev, void *data, mutex_lock(&dev->struct_mutex); - i915_kernel_lost_context(dev); obj = drm_gem_object_lookup(dev, file_priv, args->handle); if (obj == NULL) { DRM_ERROR("Bad handle in i915_gem_unpin_ioctl(): %d\n", @@ -1711,7 +1708,6 @@ int i915_gem_init_object(struct drm_gem_object *obj) void i915_gem_free_object(struct drm_gem_object *obj) { - i915_kernel_lost_context(obj->dev); i915_gem_object_unbind(obj); drm_free(obj->driver_private, 1, DRM_MEM_DRIVER); @@ -1729,7 +1725,6 @@ i915_gem_set_domain(struct drm_gem_object *obj, BUG_ON(!mutex_is_locked(&dev->struct_mutex)); drm_client_lock_take(dev, file_priv); - i915_kernel_lost_context(dev); ret = i915_gem_object_set_domain(obj, read_domains, write_domain); if (ret) { drm_client_lock_release(dev); -- cgit v1.2.3 From 2a35d857b35c9d30d073e2372aa302cd999321ba Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 11 Jun 2008 12:20:56 -0700 Subject: Remove override of drm module list in preparation for merge. --- linux-core/Makefile | 1 - 1 file changed, 1 deletion(-) diff --git a/linux-core/Makefile b/linux-core/Makefile index fc32676f..3af6f370 100644 --- a/linux-core/Makefile +++ b/linux-core/Makefile @@ -30,7 +30,6 @@ # # make DRM_MODULES="r128 radeon" # -DRM_MODULES=i915 SHELL=/bin/sh -- cgit v1.2.3 From 2655005762b8915d5f44d1d1ee7e6c2eb34841d7 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 11 Jun 2008 14:42:40 -0700 Subject: [gem] Move potentially device-specific ioctls to the intel driver. This is the create (may want location flags), pread/pwrite/mmap (performance tuning hints), and set_domain (will 32 bits be enough for everyone?) ioctls. Left in the generic set are just flink/open/close. The 2D driver must be updated for this change, and API but not ABI is broken for 3D. The driver version is bumped to mark this. --- libdrm/intel/intel_bufmgr_fake.c | 2 +- libdrm/intel/intel_bufmgr_gem.c | 30 ++--- linux-core/drmP.h | 41 +----- linux-core/drm_drv.c | 5 - linux-core/drm_gem.c | 231 +------------------------------- linux-core/i915_drv.c | 2 - linux-core/i915_gem.c | 281 ++++++++++++++++++++++++++++++++------- shared-core/drm.h | 83 +----------- shared-core/i915_dma.c | 5 + shared-core/i915_drm.h | 101 ++++++++++++-- shared-core/i915_drv.h | 20 +-- tests/gem_basic.c | 9 +- tests/gem_mmap.c | 19 +-- tests/gem_readwrite.c | 13 +- 14 files changed, 389 insertions(+), 453 deletions(-) diff --git a/libdrm/intel/intel_bufmgr_fake.c b/libdrm/intel/intel_bufmgr_fake.c index 3f5a22d3..06e85b2b 100644 --- a/libdrm/intel/intel_bufmgr_fake.c +++ b/libdrm/intel/intel_bufmgr_fake.c @@ -1062,7 +1062,7 @@ dri_fake_process_relocs(dri_bo *batch_buf) dri_fake_calculate_domains(batch_buf); - batch_fake->read_domains = DRM_GEM_DOMAIN_I915_COMMAND; + batch_fake->read_domains = I915_GEM_DOMAIN_COMMAND; /* we've ran out of RAM so blow the whole lot away and retry */ restart: diff --git a/libdrm/intel/intel_bufmgr_gem.c b/libdrm/intel/intel_bufmgr_gem.c index a65ae982..e057d949 100644 --- a/libdrm/intel/intel_bufmgr_gem.c +++ b/libdrm/intel/intel_bufmgr_gem.c @@ -299,7 +299,7 @@ dri_gem_bo_alloc(dri_bufmgr *bufmgr, const char *name, } if (!alloc_from_cache) { - struct drm_gem_create create; + struct drm_i915_gem_create create; bo_gem = calloc(1, sizeof(*bo_gem)); if (!bo_gem) @@ -309,7 +309,7 @@ dri_gem_bo_alloc(dri_bufmgr *bufmgr, const char *name, memset(&create, 0, sizeof(create)); create.size = bo_size; - ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_GEM_CREATE, &create); + ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_CREATE, &create); bo_gem->gem_handle = create.handle; if (ret != 0) { free(bo_gem); @@ -455,7 +455,7 @@ dri_gem_bo_map(dri_bo *bo, int write_enable) { dri_bufmgr_gem *bufmgr_gem; dri_bo_gem *bo_gem = (dri_bo_gem *)bo; - struct drm_gem_set_domain set_domain; + struct drm_i915_gem_set_domain set_domain; int ret; bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr; @@ -470,13 +470,13 @@ dri_gem_bo_map(dri_bo *bo, int write_enable) DBG("bo_map: %d (%s)\n", bo_gem->gem_handle, bo_gem->name); if (bo_gem->virtual == NULL) { - struct drm_gem_mmap mmap_arg; + struct drm_i915_gem_mmap mmap_arg; memset(&mmap_arg, 0, sizeof(mmap_arg)); mmap_arg.handle = bo_gem->gem_handle; mmap_arg.offset = 0; mmap_arg.size = bo->size; - ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_GEM_MMAP, &mmap_arg); + ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_MMAP, &mmap_arg); if (ret != 0) { fprintf(stderr, "%s:%d: Error mapping buffer %d (%s): %s .\n", __FILE__, __LINE__, @@ -491,9 +491,9 @@ dri_gem_bo_map(dri_bo *bo, int write_enable) if (!bo_gem->cpu_domain_set) { set_domain.handle = bo_gem->gem_handle; - set_domain.read_domains = DRM_GEM_DOMAIN_CPU; - set_domain.write_domain = write_enable ? DRM_GEM_DOMAIN_CPU : 0; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_GEM_SET_DOMAIN, &set_domain); + set_domain.read_domains = I915_GEM_DOMAIN_CPU; + set_domain.write_domain = write_enable ? I915_GEM_DOMAIN_CPU : 0; + ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain); if (ret != 0) { fprintf (stderr, "%s:%d: Error setting memory domains %d (%08x %08x): %s .\n", __FILE__, __LINE__, @@ -525,7 +525,7 @@ dri_gem_bo_subdata (dri_bo *bo, unsigned long offset, { dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr; dri_bo_gem *bo_gem = (dri_bo_gem *)bo; - struct drm_gem_pwrite pwrite; + struct drm_i915_gem_pwrite pwrite; int ret; memset (&pwrite, 0, sizeof (pwrite)); @@ -533,7 +533,7 @@ dri_gem_bo_subdata (dri_bo *bo, unsigned long offset, pwrite.offset = offset; pwrite.size = size; pwrite.data_ptr = (uint64_t) (uintptr_t) data; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_GEM_PWRITE, &pwrite); + ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PWRITE, &pwrite); if (ret != 0) { fprintf (stderr, "%s:%d: Error writing data to buffer %d: (%d %d) %s .\n", __FILE__, __LINE__, @@ -549,7 +549,7 @@ dri_gem_bo_get_subdata (dri_bo *bo, unsigned long offset, { dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr; dri_bo_gem *bo_gem = (dri_bo_gem *)bo; - struct drm_gem_pread pread; + struct drm_i915_gem_pread pread; int ret; memset (&pread, 0, sizeof (pread)); @@ -557,7 +557,7 @@ dri_gem_bo_get_subdata (dri_bo *bo, unsigned long offset, pread.offset = offset; pread.size = size; pread.data_ptr = (uint64_t) (uintptr_t) data; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_GEM_PREAD, &pread); + ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PREAD, &pread); if (ret != 0) { fprintf (stderr, "%s:%d: Error reading data from buffer %d: (%d %d) %s .\n", __FILE__, __LINE__, @@ -572,13 +572,13 @@ dri_gem_bo_wait_rendering(dri_bo *bo) { dri_bufmgr_gem *bufmgr_gem = (dri_bufmgr_gem *)bo->bufmgr; dri_bo_gem *bo_gem = (dri_bo_gem *)bo; - struct drm_gem_set_domain set_domain; + struct drm_i915_gem_set_domain set_domain; int ret; set_domain.handle = bo_gem->gem_handle; - set_domain.read_domains = DRM_GEM_DOMAIN_CPU; + set_domain.read_domains = I915_GEM_DOMAIN_CPU; set_domain.write_domain = 0; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_GEM_SET_DOMAIN, &set_domain); + ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain); if (ret != 0) { fprintf (stderr, "%s:%d: Error setting memory domains %d (%08x %08x): %s .\n", __FILE__, __LINE__, diff --git a/linux-core/drmP.h b/linux-core/drmP.h index 6a7f28d1..48ff4b87 100644 --- a/linux-core/drmP.h +++ b/linux-core/drmP.h @@ -779,21 +779,6 @@ struct drm_driver { int (*gem_init_object) (struct drm_gem_object *obj); void (*gem_free_object) (struct drm_gem_object *obj); - /** - * Driver-specific callback to set memory domains from userspace - */ - int (*gem_set_domain) (struct drm_gem_object *obj, - struct drm_file *file_priv, - uint32_t read_domains, - uint32_t write_domain); - - /** - * Driver-specific callback to flush pwrite through chipset - */ - int (*gem_flush_pwrite) (struct drm_gem_object *obj, - uint64_t offset, - uint64_t size); - struct drm_fence_driver *fence_driver; struct drm_bo_driver *bo_driver; @@ -1390,6 +1375,11 @@ static inline void drm_gem_object_unreference(struct drm_gem_object *obj) kref_put (&obj->refcount, drm_gem_object_free); } +int +drm_gem_handle_create(struct drm_file *file_priv, + struct drm_gem_object *obj, + int *handlep); + static inline void drm_gem_object_handle_reference (struct drm_gem_object *obj) { drm_gem_object_reference (obj); @@ -1413,37 +1403,16 @@ static inline void drm_gem_object_handle_unreference (struct drm_gem_object *obj struct drm_gem_object * drm_gem_object_lookup(struct drm_device *dev, struct drm_file *filp, int handle); -int drm_gem_create_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); int drm_gem_close_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); -int drm_gem_pread_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); -int drm_gem_pwrite_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); -int drm_gem_mmap_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); int drm_gem_flink_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); int drm_gem_open_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv); -int drm_gem_set_domain_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv); void drm_gem_open(struct drm_device *dev, struct drm_file *file_private); void drm_gem_release(struct drm_device *dev, struct drm_file *file_private); - -/* - * Given the new read/write domains for an object, - * compute the invalidate/flush domains for the whole device. - * - */ -int drm_gem_object_set_domain (struct drm_gem_object *object, - uint32_t read_domains, - uint32_t write_domains); - - extern void drm_core_ioremap(struct drm_map *map, struct drm_device *dev); extern void drm_core_ioremapfree(struct drm_map *map, struct drm_device *dev); diff --git a/linux-core/drm_drv.c b/linux-core/drm_drv.c index edc1f057..980752c2 100644 --- a/linux-core/drm_drv.c +++ b/linux-core/drm_drv.c @@ -151,14 +151,9 @@ static struct drm_ioctl_desc drm_ioctls[] = { DRM_IOCTL_DEF(DRM_IOCTL_MM_INFO, drm_mm_info_ioctl, 0), - DRM_IOCTL_DEF(DRM_IOCTL_GEM_CREATE, drm_gem_create_ioctl, 0), DRM_IOCTL_DEF(DRM_IOCTL_GEM_CLOSE, drm_gem_close_ioctl, 0), - DRM_IOCTL_DEF(DRM_IOCTL_GEM_PREAD, drm_gem_pread_ioctl, 0), - DRM_IOCTL_DEF(DRM_IOCTL_GEM_PWRITE, drm_gem_pwrite_ioctl, 0), - DRM_IOCTL_DEF(DRM_IOCTL_GEM_MMAP, drm_gem_mmap_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), - DRM_IOCTL_DEF(DRM_IOCTL_GEM_SET_DOMAIN, drm_gem_set_domain_ioctl, DRM_AUTH), }; #define DRM_CORE_IOCTL_COUNT ARRAY_SIZE( drm_ioctls ) diff --git a/linux-core/drm_gem.c b/linux-core/drm_gem.c index a8ecaf76..47b55434 100644 --- a/linux-core/drm_gem.c +++ b/linux-core/drm_gem.c @@ -99,15 +99,6 @@ drm_gem_object_alloc(struct drm_device *dev, size_t size) kref_init(&obj->refcount); kref_init(&obj->handlecount); obj->size = size; - - /* - * 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 = DRM_GEM_DOMAIN_CPU; - obj->read_domains = DRM_GEM_DOMAIN_CPU; if (dev->driver->gem_init_object != NULL && dev->driver->gem_init_object(obj) != 0) { fput(obj->filp); @@ -163,7 +154,7 @@ drm_gem_handle_delete(struct drm_file *filp, int handle) * to the object, which includes a regular reference count. Callers * will likely want to dereference the object afterwards. */ -static int +int drm_gem_handle_create(struct drm_file *file_priv, struct drm_gem_object *obj, int *handlep) @@ -191,6 +182,7 @@ again: 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 * @@ -216,40 +208,6 @@ drm_gem_object_lookup(struct drm_device *dev, struct drm_file *filp, } EXPORT_SYMBOL(drm_gem_object_lookup); -/** - * Creates a new mm object and returns a handle to it. - */ -int -drm_gem_create_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_gem_create *args = data; - struct drm_gem_object *obj; - int handle, ret; - - if (!(dev->driver->driver_features & DRIVER_GEM)) - return -ENODEV; - - 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; -} - /** * Releases the handle to an mm object. */ @@ -268,158 +226,6 @@ drm_gem_close_ioctl(struct drm_device *dev, void *data, return ret; } -/** - * Reads data from the object referenced by handle. - * - * On error, the contents of *data are undefined. - */ -int -drm_gem_pread_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_gem_pread *args = data; - struct drm_gem_object *obj; - ssize_t read; - loff_t offset; - 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; - - mutex_lock(&dev->struct_mutex); - if (dev->driver->gem_set_domain) { - ret = dev->driver->gem_set_domain(obj, file_priv, - DRM_GEM_DOMAIN_CPU, - 0); - if (ret) { - drm_gem_object_unreference(obj); - mutex_unlock(&dev->struct_mutex); - return ret; - } - } - 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; -} - -/** - * 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 -drm_gem_mmap_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_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 -EINVAL; - - 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; -} - -/** - * Writes data to the object referenced by handle. - * - * On error, the contents of the buffer that were to be modified are undefined. - */ -int -drm_gem_pwrite_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_gem_pwrite *args = data; - struct drm_gem_object *obj; - ssize_t written; - loff_t offset; - 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; - - mutex_lock(&dev->struct_mutex); - if (dev->driver->gem_set_domain) { - ret = dev->driver->gem_set_domain(obj, file_priv, - DRM_GEM_DOMAIN_CPU, - DRM_GEM_DOMAIN_CPU); - if (ret) { - drm_gem_object_unreference(obj); - 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) { - drm_gem_object_unreference(obj); - mutex_unlock(&dev->struct_mutex); - if (written < 0) - return written; - else - return -EINVAL; - } - - if (dev->driver->gem_flush_pwrite) - dev->driver->gem_flush_pwrite(obj, - args->offset, - args->size); - - drm_gem_object_unreference(obj); - mutex_unlock(&dev->struct_mutex); - - return 0; -} - /** * Create a global name for an object, returning the name. * @@ -511,39 +317,6 @@ drm_gem_open_ioctl(struct drm_device *dev, void *data, return 0; } -/** - * Called when user space prepares to use an object - */ -int -drm_gem_set_domain_ioctl(struct drm_device *dev, void *data, - struct drm_file *file_priv) -{ - struct drm_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 -EINVAL; - - mutex_lock(&dev->struct_mutex); - if (dev->driver->gem_set_domain) { - ret = dev->driver->gem_set_domain(obj, file_priv, - args->read_domains, - args->write_domain); - } else { - obj->read_domains = args->read_domains; - obj->write_domain = args->write_domain; - ret = 0; - } - drm_gem_object_unreference(obj); - mutex_unlock(&dev->struct_mutex); - return ret; -} - /** * Called at device open time, sets up the structure for handling refcounting * of mm objects. diff --git a/linux-core/i915_drv.c b/linux-core/i915_drv.c index 3f246a08..8f51cd4d 100644 --- a/linux-core/i915_drv.c +++ b/linux-core/i915_drv.c @@ -588,8 +588,6 @@ static struct drm_driver driver = { .ioctls = i915_ioctls, .gem_init_object = i915_gem_init_object, .gem_free_object = i915_gem_free_object, - .gem_set_domain = i915_gem_set_domain, - .gem_flush_pwrite = i915_gem_flush_pwrite, .fops = { .owner = THIS_MODULE, .open = drm_open, diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 3d47ec1a..e51de316 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -40,6 +40,11 @@ static int i915_gem_object_set_domain(struct drm_gem_object *obj, uint32_t read_domains, uint32_t write_domain); +int +i915_gem_set_domain(struct drm_gem_object *obj, + struct drm_file *file_priv, + uint32_t read_domains, + uint32_t write_domain); int i915_gem_init_ioctl(struct drm_device *dev, void *data, @@ -65,6 +70,199 @@ i915_gem_init_ioctl(struct drm_device *dev, void *data, 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; + ssize_t read; + loff_t offset; + int ret; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + + mutex_lock(&dev->struct_mutex); + ret = i915_gem_set_domain(obj, file_priv, + I915_GEM_DOMAIN_CPU, 0); + if (ret) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + 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; +} + +/** + * 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; + ssize_t written; + loff_t offset; + int ret; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + + mutex_lock(&dev->struct_mutex); + ret = i915_gem_set_domain(obj, file_priv, + I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU); + if (ret) { + drm_gem_object_unreference(obj); + 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) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (written < 0) + return written; + else + return -EINVAL; + } + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +/** + * 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 -EINVAL; + + mutex_lock(&dev->struct_mutex); + 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; +} + +/** + * 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 -EINVAL; + + 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) { @@ -187,7 +385,7 @@ i915_retire_commands(struct drm_device *dev) /* The sampler always gets flushed on i965 (sigh) */ if (IS_I965G(dev)) - flush_domains |= DRM_GEM_DOMAIN_I915_SAMPLER; + flush_domains |= I915_GEM_DOMAIN_SAMPLER; BEGIN_LP_RING(2); OUT_RING(cmd); OUT_RING(0); /* noop */ @@ -377,51 +575,51 @@ i915_gem_flush(struct drm_device *dev, invalidate_domains, flush_domains); #endif - if (flush_domains & DRM_GEM_DOMAIN_CPU) + if (flush_domains & I915_GEM_DOMAIN_CPU) drm_agp_chipset_flush(dev); - if ((invalidate_domains|flush_domains) & ~DRM_GEM_DOMAIN_CPU) { + if ((invalidate_domains|flush_domains) & ~I915_GEM_DOMAIN_CPU) { /* * read/write caches: * - * DRM_GEM_DOMAIN_I915_RENDER is always invalidated, but is + * 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: * - * DRM_GEM_DOMAIN_I915_SAMPLER is flushed on pre-965 if + * I915_GEM_DOMAIN_SAMPLER is flushed on pre-965 if * MI_READ_FLUSH is set, and is always flushed on 965. * - * DRM_GEM_DOMAIN_I915_COMMAND may not exist? + * I915_GEM_DOMAIN_COMMAND may not exist? * - * DRM_GEM_DOMAIN_I915_INSTRUCTION, which exists on 965, is + * I915_GEM_DOMAIN_INSTRUCTION, which exists on 965, is * invalidated when MI_EXE_FLUSH is set. * - * DRM_GEM_DOMAIN_I915_VERTEX, which exists on 965, is + * I915_GEM_DOMAIN_VERTEX, which exists on 965, is * invalidated with every MI_FLUSH. * * TLBs: * - * On 965, TLBs associated with DRM_GEM_DOMAIN_I915_COMMAND - * and DRM_GEM_DOMAIN_CPU in are invalidated at PTE write and - * DRM_GEM_DOMAIN_I915_RENDER and DRM_GEM_DOMAIN_I915_SAMPLER + * 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 = CMD_MI_FLUSH | MI_NO_WRITE_FLUSH; if ((invalidate_domains|flush_domains) & - DRM_GEM_DOMAIN_I915_RENDER) + 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 & DRM_GEM_DOMAIN_I915_SAMPLER) + if (invalidate_domains & I915_GEM_DOMAIN_SAMPLER) cmd |= MI_READ_FLUSH; } - if (invalidate_domains & DRM_GEM_DOMAIN_I915_INSTRUCTION) + if (invalidate_domains & I915_GEM_DOMAIN_INSTRUCTION) cmd |= MI_EXE_FLUSH; #if WATCH_EXEC @@ -448,7 +646,7 @@ i915_gem_object_wait_rendering(struct drm_gem_object *obj) /* If there are writes queued to the buffer, flush and * create a new seqno to wait for. */ - if (obj->write_domain & ~(DRM_GEM_DOMAIN_CPU)) { + if (obj->write_domain & ~(I915_GEM_DOMAIN_CPU)) { uint32_t write_domain = obj->write_domain; #if WATCH_BUF DRM_INFO("%s: flushing object %p from write domain %08x\n", @@ -503,8 +701,8 @@ i915_gem_object_unbind(struct drm_gem_object *obj) * also ensure that all pending GPU writes are finished * before we unbind. */ - ret = i915_gem_object_set_domain (obj, DRM_GEM_DOMAIN_CPU, - DRM_GEM_DOMAIN_CPU); + ret = i915_gem_object_set_domain (obj, I915_GEM_DOMAIN_CPU, + I915_GEM_DOMAIN_CPU); if (ret) return ret; @@ -805,8 +1003,8 @@ i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment) * wasn't in the GTT, there shouldn't be any way it could have been in * a GPU cache */ - BUG_ON(obj->read_domains & ~DRM_GEM_DOMAIN_CPU); - BUG_ON(obj->write_domain & ~DRM_GEM_DOMAIN_CPU); + BUG_ON(obj->read_domains & ~I915_GEM_DOMAIN_CPU); + BUG_ON(obj->write_domain & ~I915_GEM_DOMAIN_CPU); return 0; } @@ -973,7 +1171,7 @@ i915_gem_object_set_domain(struct drm_gem_object *obj, * stale data. That is, any new read domains. */ invalidate_domains |= read_domains & ~obj->read_domains; - if ((flush_domains | invalidate_domains) & DRM_GEM_DOMAIN_CPU) { + 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); @@ -983,8 +1181,8 @@ i915_gem_object_set_domain(struct drm_gem_object *obj, * then pause for rendering so that the GPU caches will be * flushed before the cpu cache is invalidated */ - if ((invalidate_domains & DRM_GEM_DOMAIN_CPU) && - (flush_domains & ~DRM_GEM_DOMAIN_CPU)) { + if ((invalidate_domains & I915_GEM_DOMAIN_CPU) && + (flush_domains & ~I915_GEM_DOMAIN_CPU)) { ret = i915_gem_object_wait_rendering(obj); if (ret) return ret; @@ -1225,7 +1423,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, /* As we're writing through the gtt, flush * any CPU writes before we write the relocations */ - if (obj->write_domain & DRM_GEM_DOMAIN_CPU) { + if (obj->write_domain & I915_GEM_DOMAIN_CPU) { i915_gem_clflush_object(obj); drm_agp_chipset_flush(dev); obj->write_domain = 0; @@ -1466,7 +1664,7 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, /* Set the pending read domains for the batch buffer to COMMAND */ batch_obj = object_list[args->buffer_count-1]; - batch_obj->pending_read_domains = DRM_GEM_DOMAIN_I915_COMMAND; + batch_obj->pending_read_domains = I915_GEM_DOMAIN_COMMAND; batch_obj->pending_write_domain = 0; for (i = 0; i < args->buffer_count; i++) { @@ -1700,6 +1898,15 @@ int i915_gem_init_object(struct drm_gem_object *obj) 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); @@ -1735,32 +1942,6 @@ i915_gem_set_domain(struct drm_gem_object *obj, return 0; } -int -i915_gem_flush_pwrite(struct drm_gem_object *obj, - uint64_t offset, uint64_t size) -{ -#if 0 - struct drm_device *dev = obj->dev; - struct drm_i915_gem_object *obj_priv = obj->driver_private; - - /* - * For writes much less than the size of the object and - * which are already pinned in memory, do the flush right now - */ - - if ((size < obj->size >> 1) && obj_priv->page_list != NULL) { - unsigned long first_page = offset / PAGE_SIZE; - unsigned long beyond_page = roundup(offset + size, PAGE_SIZE) / PAGE_SIZE; - - drm_ttm_cache_flush(obj_priv->page_list + first_page, - beyond_page - first_page); - drm_agp_chipset_flush(dev); - obj->write_domain = 0; - } -#endif - return 0; -} - void i915_gem_lastclose(struct drm_device *dev) { diff --git a/shared-core/drm.h b/shared-core/drm.h index 2373a22e..6012ef21 100644 --- a/shared-core/drm.h +++ b/shared-core/drm.h @@ -959,69 +959,12 @@ struct drm_mm_info_arg { uint64_t p_size; }; - -struct drm_gem_create { - /** - * Requested size for the object. - * - * The (page-aligned) allocated size for the object will be returned. - */ - uint64_t size; - /** - * Returned handle for the object. - * - * Object handles are nonzero. - */ - uint32_t handle; - uint32_t pad; -}; - struct drm_gem_close { /** Handle of the object to be closed. */ uint32_t handle; uint32_t pad; }; -struct drm_gem_pread { - /** Handle for the object being read. */ - uint32_t handle; - uint32_t pad; - /** Offset into the object to read from */ - uint64_t offset; - /** Length of data to read */ - uint64_t size; - /** Pointer to write the data into. */ - uint64_t data_ptr; /* void *, but pointers are not 32/64 compatible */ -}; - -struct drm_gem_pwrite { - /** Handle for the object being written to. */ - uint32_t handle; - uint32_t pad; - /** Offset into the object to write to */ - uint64_t offset; - /** Length of data to write */ - uint64_t size; - /** Pointer to read the data from. */ - uint64_t data_ptr; /* void *, but pointers are not 32/64 compatible */ -}; - -struct drm_gem_mmap { - /** Handle for the object being mapped. */ - uint32_t handle; - uint32_t pad; - /** Offset in the object to map. */ - uint64_t offset; - /** - * Length of data to map. - * - * The value will be page-aligned. - */ - uint64_t size; - /** Returned pointer the data was mapped at */ - uint64_t addr_ptr; /* void *, but pointers are not 32/64 compatible */ -}; - struct drm_gem_flink { /** Handle for the object being named */ uint32_t handle; @@ -1041,19 +984,6 @@ struct drm_gem_open { uint64_t size; }; -struct drm_gem_set_domain { - /** Handle for the object */ - uint32_t handle; - - /** New read domains */ - uint32_t read_domains; - - /** New write domain */ - uint32_t write_domain; -}; - -#define DRM_GEM_DOMAIN_CPU 0x00000001 - /** * \name Ioctls Definitions */ @@ -1075,6 +1005,10 @@ struct drm_gem_set_domain { #define DRM_IOCTL_SET_VERSION DRM_IOWR(0x07, struct drm_set_version) #define DRM_IOCTL_MODESET_CTL DRM_IOW(0x08, struct drm_modeset_ctl) +#define DRM_IOCTL_GEM_CLOSE DRM_IOW (0x09, struct drm_gem_close) +#define DRM_IOCTL_GEM_FLINK DRM_IOWR(0x0a, struct drm_gem_flink) +#define DRM_IOCTL_GEM_OPEN DRM_IOWR(0x0b, struct drm_gem_open) + #define DRM_IOCTL_SET_UNIQUE DRM_IOW( 0x10, struct drm_unique) #define DRM_IOCTL_AUTH_MAGIC DRM_IOW( 0x11, struct drm_auth) #define DRM_IOCTL_BLOCK DRM_IOWR(0x12, struct drm_block) @@ -1122,15 +1056,6 @@ struct drm_gem_set_domain { #define DRM_IOCTL_UPDATE_DRAW DRM_IOW(0x3f, struct drm_update_draw) -#define DRM_IOCTL_GEM_CREATE DRM_IOWR(0x09, struct drm_gem_create) -#define DRM_IOCTL_GEM_CLOSE DRM_IOW (0x0a, struct drm_gem_close) -#define DRM_IOCTL_GEM_PREAD DRM_IOW (0x0b, struct drm_gem_pread) -#define DRM_IOCTL_GEM_PWRITE DRM_IOW (0x0c, struct drm_gem_pwrite) -#define DRM_IOCTL_GEM_MMAP DRM_IOWR(0x0d, struct drm_gem_mmap) -#define DRM_IOCTL_GEM_FLINK DRM_IOWR(0x0e, struct drm_gem_flink) -#define DRM_IOCTL_GEM_OPEN DRM_IOWR(0x0f, struct drm_gem_open) -#define DRM_IOCTL_GEM_SET_DOMAIN DRM_IOW (0xb7, struct drm_gem_set_domain) - #define DRM_IOCTL_MM_INIT DRM_IOWR(0xc0, struct drm_mm_init_arg) #define DRM_IOCTL_MM_TAKEDOWN DRM_IOWR(0xc1, struct drm_mm_type_arg) #define DRM_IOCTL_MM_LOCK DRM_IOWR(0xc2, struct drm_mm_type_arg) diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c index 4243b4e9..7e4de18b 100644 --- a/shared-core/i915_dma.c +++ b/shared-core/i915_dma.c @@ -1190,6 +1190,11 @@ struct drm_ioctl_desc i915_ioctls[] = { 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), }; int i915_max_ioctl = DRM_ARRAY_SIZE(i915_ioctls); diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h index ce01640a..fe47708d 100644 --- a/shared-core/i915_drm.h +++ b/shared-core/i915_drm.h @@ -183,7 +183,12 @@ typedef struct drm_i915_sarea { #define DRM_I915_GEM_BUSY 0x17 #define DRM_I915_GEM_THROTTLE 0x18 #define DRM_I915_GEM_ENTERVT 0x19 -#define DRM_I915_GEM_LEAVEVT 0x20 +#define DRM_I915_GEM_LEAVEVT 0x1a +#define DRM_I915_GEM_CREATE 0x1b +#define DRM_I915_GEM_PREAD 0x1c +#define DRM_I915_GEM_PWRITE 0x1d +#define DRM_I915_GEM_MMAP 0x1e +#define DRM_I915_GEM_SET_DOMAIN 0x1f #define DRM_IOCTL_I915_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t) #define DRM_IOCTL_I915_FLUSH DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH) @@ -211,6 +216,11 @@ typedef struct drm_i915_sarea { #define DRM_IOCTL_I915_GEM_THROTTLE DRM_IO ( DRM_COMMAND_BASE + DRM_I915_GEM_THROTTLE) #define DRM_IOCTL_I915_GEM_ENTERVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_ENTERVT) #define DRM_IOCTL_I915_GEM_LEAVEVT DRM_IO(DRM_COMMAND_BASE + DRM_I915_GEM_LEAVEVT) +#define DRM_IOCTL_I915_GEM_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_CREATE, struct drm_i915_gem_create) +#define DRM_IOCTL_I915_GEM_PREAD DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PREAD, struct drm_i915_gem_pread) +#define DRM_IOCTL_I915_GEM_PWRITE DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PWRITE, struct drm_i915_gem_pwrite) +#define DRM_IOCTL_I915_GEM_MMAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP, struct drm_i915_gem_mmap) +#define DRM_IOCTL_I915_GEM_SET_DOMAIN DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SET_DOMAIN, struct drm_i915_gem_set_domain) /* Asynchronous page flipping: */ @@ -424,6 +434,73 @@ struct drm_i915_gem_init { uint64_t gtt_end; }; +struct drm_i915_gem_create { + /** + * Requested size for the object. + * + * The (page-aligned) allocated size for the object will be returned. + */ + uint64_t size; + /** + * Returned handle for the object. + * + * Object handles are nonzero. + */ + uint32_t handle; + uint32_t pad; +}; + +struct drm_i915_gem_pread { + /** Handle for the object being read. */ + uint32_t handle; + uint32_t pad; + /** Offset into the object to read from */ + uint64_t offset; + /** Length of data to read */ + uint64_t size; + /** Pointer to write the data into. */ + uint64_t data_ptr; /* void *, but pointers are not 32/64 compatible */ +}; + +struct drm_i915_gem_pwrite { + /** Handle for the object being written to. */ + uint32_t handle; + uint32_t pad; + /** Offset into the object to write to */ + uint64_t offset; + /** Length of data to write */ + uint64_t size; + /** Pointer to read the data from. */ + uint64_t data_ptr; /* void *, but pointers are not 32/64 compatible */ +}; + +struct drm_i915_gem_mmap { + /** Handle for the object being mapped. */ + uint32_t handle; + uint32_t pad; + /** Offset in the object to map. */ + uint64_t offset; + /** + * Length of data to map. + * + * The value will be page-aligned. + */ + uint64_t size; + /** Returned pointer the data was mapped at */ + uint64_t addr_ptr; /* void *, but pointers are not 32/64 compatible */ +}; + +struct drm_i915_gem_set_domain { + /** Handle for the object */ + uint32_t handle; + + /** New read domains */ + uint32_t read_domains; + + /** New write domain */ + uint32_t write_domain; +}; + struct drm_i915_gem_relocation_entry { /** * Handle of the buffer being pointed to by this relocation entry. @@ -469,20 +546,26 @@ struct drm_i915_gem_relocation_entry { uint32_t write_domain; }; -/** +/** @{ * Intel memory domains * * Most of these just align with the various caches in * the system and are used to flush and invalidate as * objects end up cached in different domains. */ - -/* 0x00000001 is DRM_GEM_DOMAIN_CPU */ -#define DRM_GEM_DOMAIN_I915_RENDER 0x00000002 /* Render cache, used by 2D and 3D drawing */ -#define DRM_GEM_DOMAIN_I915_SAMPLER 0x00000004 /* Sampler cache, used by texture engine */ -#define DRM_GEM_DOMAIN_I915_COMMAND 0x00000008 /* Command queue, used to load batch buffers */ -#define DRM_GEM_DOMAIN_I915_INSTRUCTION 0x00000010 /* Instruction cache, used by shader programs */ -#define DRM_GEM_DOMAIN_I915_VERTEX 0x00000020 /* Vertex address cache */ +/** CPU cache */ +#define I915_GEM_DOMAIN_CPU 0x00000001 +/** Render cache, used by 2D and 3D drawing */ +#define I915_GEM_DOMAIN_RENDER 0x00000002 +/** Sampler cache, used by texture engine */ +#define I915_GEM_DOMAIN_SAMPLER 0x00000004 +/** Command queue, used to load batch buffers */ +#define I915_GEM_DOMAIN_COMMAND 0x00000008 +/** Instruction cache, used by shader programs */ +#define I915_GEM_DOMAIN_INSTRUCTION 0x00000010 +/** Vertex address cache */ +#define I915_GEM_DOMAIN_VERTEX 0x00000020 +/** @} */ struct drm_i915_gem_exec_object { /** diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h index 3a22ae3c..9c9925b5 100644 --- a/shared-core/i915_drv.h +++ b/shared-core/i915_drv.h @@ -37,7 +37,7 @@ #define DRIVER_NAME "i915" #define DRIVER_DESC "Intel Graphics" -#define DRIVER_DATE "20080312" +#define DRIVER_DATE "20080611" #if defined(__linux__) #define I915_HAVE_FENCE @@ -61,7 +61,7 @@ */ #define DRIVER_MAJOR 1 #if defined(I915_HAVE_FENCE) && defined(I915_HAVE_BUFFER) -#define DRIVER_MINOR 13 +#define DRIVER_MINOR 14 #else #define DRIVER_MINOR 6 #endif @@ -461,6 +461,16 @@ int i915_execbuffer(struct drm_device *dev, void *data, /* 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_execbuffer(struct drm_device *dev, void *data, struct drm_file *file_priv); int i915_gem_pin_ioctl(struct drm_device *dev, void *data, @@ -479,12 +489,6 @@ 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); -int i915_gem_set_domain(struct drm_gem_object *obj, - struct drm_file *file_priv, - uint32_t read_domains, - uint32_t write_domain); -int i915_gem_flush_pwrite(struct drm_gem_object *obj, - uint64_t offset, uint64_t size); void i915_gem_lastclose(struct drm_device *dev); void i915_gem_retire_requests(struct drm_device *dev); void i915_gem_retire_timeout(unsigned long data); diff --git a/tests/gem_basic.c b/tests/gem_basic.c index 8b8b63d0..b2176fba 100644 --- a/tests/gem_basic.c +++ b/tests/gem_basic.c @@ -34,6 +34,7 @@ #include #include #include "drm.h" +#include "i915_drm.h" static void test_bad_close(int fd) @@ -52,7 +53,7 @@ test_bad_close(int fd) static void test_create_close(int fd) { - struct drm_gem_create create; + struct drm_i915_gem_create create; struct drm_gem_close close; int ret; @@ -60,7 +61,7 @@ test_create_close(int fd) memset(&create, 0, sizeof(create)); create.size = 16 * 1024; - ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + ret = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE, &create); assert(ret == 0); close.handle = create.handle; @@ -70,14 +71,14 @@ test_create_close(int fd) static void test_create_fd_close(int fd) { - struct drm_gem_create create; + struct drm_i915_gem_create create; int ret; printf("Testing closing with an object allocated.\n"); memset(&create, 0, sizeof(create)); create.size = 16 * 1024; - ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + ret = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE, &create); assert(ret == 0); close(fd); diff --git a/tests/gem_mmap.c b/tests/gem_mmap.c index 3f8e27a0..c3a51883 100644 --- a/tests/gem_mmap.c +++ b/tests/gem_mmap.c @@ -34,12 +34,13 @@ #include #include #include "drm.h" +#include "i915_drm.h" #define OBJECT_SIZE 16384 int do_read(int fd, int handle, void *buf, int offset, int size) { - struct drm_gem_pread read; + struct drm_i915_gem_pread read; /* Ensure that we don't have any convenient data in buf in case * we fail. @@ -52,12 +53,12 @@ int do_read(int fd, int handle, void *buf, int offset, int size) read.size = size; read.offset = offset; - return ioctl(fd, DRM_IOCTL_GEM_PREAD, &read); + return ioctl(fd, DRM_IOCTL_I915_GEM_PREAD, &read); } int do_write(int fd, int handle, void *buf, int offset, int size) { - struct drm_gem_pwrite write; + struct drm_i915_gem_pwrite write; memset(&write, 0, sizeof(write)); write.handle = handle; @@ -65,14 +66,14 @@ int do_write(int fd, int handle, void *buf, int offset, int size) write.size = size; write.offset = offset; - return ioctl(fd, DRM_IOCTL_GEM_PWRITE, &write); + return ioctl(fd, DRM_IOCTL_I915_GEM_PWRITE, &write); } int main(int argc, char **argv) { int fd; - struct drm_gem_create create; - struct drm_gem_mmap mmap; + struct drm_i915_gem_create create; + struct drm_i915_gem_mmap mmap; struct drm_gem_close unref; uint8_t expected[OBJECT_SIZE]; uint8_t buf[OBJECT_SIZE]; @@ -87,12 +88,12 @@ int main(int argc, char **argv) mmap.offset = 0; mmap.size = 4096; printf("Testing mmaping of bad object.\n"); - ret = ioctl(fd, DRM_IOCTL_GEM_MMAP, &mmap); + ret = ioctl(fd, DRM_IOCTL_I915_GEM_MMAP, &mmap); assert(ret == -1 && errno == EINVAL); memset(&create, 0, sizeof(create)); create.size = OBJECT_SIZE; - ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + ret = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE, &create); assert(ret == 0); handle = create.handle; @@ -100,7 +101,7 @@ int main(int argc, char **argv) mmap.handle = handle; mmap.offset = 0; mmap.size = OBJECT_SIZE; - ret = ioctl(fd, DRM_IOCTL_GEM_MMAP, &mmap); + ret = ioctl(fd, DRM_IOCTL_I915_GEM_MMAP, &mmap); assert(ret == 0); addr = (uint8_t *)(uintptr_t)mmap.addr_ptr; diff --git a/tests/gem_readwrite.c b/tests/gem_readwrite.c index a48f9847..54b25ea3 100644 --- a/tests/gem_readwrite.c +++ b/tests/gem_readwrite.c @@ -34,12 +34,13 @@ #include #include #include "drm.h" +#include "i915_drm.h" #define OBJECT_SIZE 16384 int do_read(int fd, int handle, void *buf, int offset, int size) { - struct drm_gem_pread read; + struct drm_i915_gem_pread read; /* Ensure that we don't have any convenient data in buf in case * we fail. @@ -52,12 +53,12 @@ int do_read(int fd, int handle, void *buf, int offset, int size) read.size = size; read.offset = offset; - return ioctl(fd, DRM_IOCTL_GEM_PREAD, &read); + return ioctl(fd, DRM_IOCTL_I915_GEM_PREAD, &read); } int do_write(int fd, int handle, void *buf, int offset, int size) { - struct drm_gem_pwrite write; + struct drm_i915_gem_pwrite write; memset(&write, 0, sizeof(write)); write.handle = handle; @@ -65,13 +66,13 @@ int do_write(int fd, int handle, void *buf, int offset, int size) write.size = size; write.offset = offset; - return ioctl(fd, DRM_IOCTL_GEM_PWRITE, &write); + return ioctl(fd, DRM_IOCTL_I915_GEM_PWRITE, &write); } int main(int argc, char **argv) { int fd; - struct drm_gem_create create; + struct drm_i915_gem_create create; uint8_t expected[OBJECT_SIZE]; uint8_t buf[OBJECT_SIZE]; int ret; @@ -81,7 +82,7 @@ int main(int argc, char **argv) memset(&create, 0, sizeof(create)); create.size = OBJECT_SIZE; - ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + ret = ioctl(fd, DRM_IOCTL_I915_GEM_CREATE, &create); assert(ret == 0); handle = create.handle; -- cgit v1.2.3 From 846d792ac10c4b2738bb5ff59e56df168b9921ff Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 11 Jun 2008 15:51:17 -0700 Subject: [gem] Another round of cleanups from checkpatch.pl --- linux-core/i915_gem.c | 19 +++++++++---------- shared-core/i915_drm.h | 20 +++++++++++++------- 2 files changed, 22 insertions(+), 17 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index e51de316..d608cb03 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -547,8 +547,8 @@ i915_wait_request(struct drm_device *dev, uint32_t seqno) i915_user_irq_off(dev_priv); } if (ret) - DRM_ERROR ("%s returns %d (awaiting %d at %d)\n", - __func__, ret, seqno, i915_get_gem_seqno(dev)); + 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 @@ -701,8 +701,8 @@ i915_gem_object_unbind(struct drm_gem_object *obj) * 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); + ret = i915_gem_object_set_domain(obj, I915_GEM_DOMAIN_CPU, + I915_GEM_DOMAIN_CPU); if (ret) return ret; @@ -1242,7 +1242,7 @@ i915_gem_object_check_coherency(struct drm_gem_object *obj, int handle) int bad_count = 0; DRM_INFO("%s: checking coherency of object %p@0x%08x (%d, %dkb):\n", - __FUNCTION__, obj, obj_priv->gtt_offset, handle, + __func__, obj, obj_priv->gtt_offset, handle, obj->size / 1024); gtt_mapping = ioremap(dev->agp->base + obj_priv->gtt_offset, @@ -1301,9 +1301,7 @@ i915_gem_object_check_coherency(struct drm_gem_object *obj, int handle) #endif /** - * Bind an object to the GTT and evaluate the relocations landing in it - * - * + * Bind an object to the GTT and evaluate the relocations landing in it. */ static int i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, @@ -1567,7 +1565,8 @@ i915_gem_ring_throttle(struct drm_device *dev) list); /* Break out if we're close enough. */ - if ((long) (jiffies - request->emitted_jiffies) <= (20 * HZ) / 1000) { + if ((long) (jiffies - request->emitted_jiffies) <= + (20 * HZ) / 1000) { mutex_unlock(&dev->struct_mutex); return 0; } @@ -1877,7 +1876,7 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data, obj_priv = obj->driver_private; args->busy = obj_priv->active; - + drm_gem_object_unreference(obj); mutex_unlock(&dev->struct_mutex); return 0; diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h index fe47708d..f16dc99f 100644 --- a/shared-core/i915_drm.h +++ b/shared-core/i915_drm.h @@ -573,11 +573,15 @@ struct drm_i915_gem_exec_object { * operation. */ uint32_t handle; - - /** List of relocations to be performed on this buffer */ + + /** Number of relocations to be performed on this buffer */ uint32_t relocation_count; - uint64_t relocs_ptr; /* struct drm_i915_gem_relocation_entry *relocs */ - + /** + * Pointer to array of struct drm_i915_gem_relocation_entry containing + * the relocations to be performed in this buffer. + */ + uint64_t relocs_ptr; + /** Required alignment in graphics aperture */ uint64_t alignment; @@ -593,11 +597,13 @@ struct drm_i915_gem_execbuffer { * List of buffers to be validated with their relocations to be * performend on them. * + * This is a pointer to an array of struct drm_i915_gem_validate_entry. + * * These buffers must be listed in an order such that all relocations * a buffer is performing refer to buffers that have already appeared * in the validate list. */ - uint64_t buffers_ptr; /* struct drm_i915_gem_validate_entry *buffers */ + uint64_t buffers_ptr; uint32_t buffer_count; /** Offset in the batchbuffer to start execution from. */ @@ -614,7 +620,7 @@ struct drm_i915_gem_pin { /** Handle of the buffer to be pinned. */ uint32_t handle; uint32_t pad; - + /** alignment required within the aperture */ uint64_t alignment; @@ -631,7 +637,7 @@ struct drm_i915_gem_unpin { struct drm_i915_gem_busy { /** Handle of the buffer to check for busy */ uint32_t handle; - + /** Return busy status (1 if busy, 0 if idle) */ uint32_t busy; }; -- cgit v1.2.3 From 57b4c4c32d3b474939775042bb8a17423def446d Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 11 Jun 2008 15:58:33 -0700 Subject: Move the renaming of mm.c symbols to #defines in the header. This reduces the diff from Mesa and reduces the illegibility of what I did. --- libdrm/intel/intel_bufmgr_fake.c | 12 ++++++------ libdrm/intel/mm.c | 12 ++++++------ libdrm/intel/mm.h | 22 +++++++++++++++------- 3 files changed, 27 insertions(+), 19 deletions(-) diff --git a/libdrm/intel/intel_bufmgr_fake.c b/libdrm/intel/intel_bufmgr_fake.c index 06e85b2b..e988eb58 100644 --- a/libdrm/intel/intel_bufmgr_fake.c +++ b/libdrm/intel/intel_bufmgr_fake.c @@ -252,7 +252,7 @@ alloc_block(dri_bo *bo) sz = (bo->size + bo_fake->alignment - 1) & ~(bo_fake->alignment - 1); - block->mem = drmmmAllocMem(bufmgr_fake->heap, sz, align_log2, 0); + block->mem = mmAllocMem(bufmgr_fake->heap, sz, align_log2, 0); if (!block->mem) { free(block); return 0; @@ -300,7 +300,7 @@ static void free_block(dri_bufmgr_fake *bufmgr_fake, struct block *block) DBG(" - free immediately\n"); DRMLISTDEL(block); - drmmmFreeMem(block->mem); + mmFreeMem(block->mem); free(block); } } @@ -415,7 +415,7 @@ static int clear_fenced(dri_bufmgr_fake *bufmgr_fake, DBG("delayed free: offset %x sz %x\n", block->mem->ofs, block->mem->size); DRMLISTDEL(block); - drmmmFreeMem(block->mem); + mmFreeMem(block->mem); free(block); } else { @@ -923,7 +923,7 @@ dri_fake_destroy(dri_bufmgr *bufmgr) { dri_bufmgr_fake *bufmgr_fake = (dri_bufmgr_fake *)bufmgr; - drmmmDestroy(bufmgr_fake->heap); + mmDestroy(bufmgr_fake->heap); free(bufmgr); } @@ -1074,7 +1074,7 @@ dri_fake_process_relocs(dri_bo *batch_buf) bufmgr_fake->fail = 0; goto restart; } else /* dump out the memory here */ - drmmmDumpMemInfo(bufmgr_fake->heap); + mmDumpMemInfo(bufmgr_fake->heap); } assert(ret == 0); @@ -1193,7 +1193,7 @@ intel_bufmgr_fake_init(unsigned long low_offset, void *low_virtual, bufmgr_fake->low_offset = low_offset; bufmgr_fake->virtual = low_virtual; bufmgr_fake->size = size; - bufmgr_fake->heap = drmmmInit(low_offset, size); + bufmgr_fake->heap = mmInit(low_offset, size); /* Hook in methods */ bufmgr_fake->bufmgr.bo_alloc = dri_fake_bo_alloc; diff --git a/libdrm/intel/mm.c b/libdrm/intel/mm.c index 2605d8ec..98146405 100644 --- a/libdrm/intel/mm.c +++ b/libdrm/intel/mm.c @@ -29,7 +29,7 @@ #include "mm.h" void -drmmmDumpMemInfo(const struct mem_block *heap) +mmDumpMemInfo(const struct mem_block *heap) { drmMsg("Memory heap %p:\n", (void *)heap); if (heap == 0) { @@ -56,7 +56,7 @@ drmmmDumpMemInfo(const struct mem_block *heap) } struct mem_block * -drmmmInit(int ofs, int size) +mmInit(int ofs, int size) { struct mem_block *heap, *block; @@ -163,7 +163,7 @@ SliceBlock(struct mem_block *p, struct mem_block * -drmmmAllocMem(struct mem_block *heap, int size, int align2, int startSearch) +mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch) { struct mem_block *p; const int mask = (1 << align2)-1; @@ -196,7 +196,7 @@ drmmmAllocMem(struct mem_block *heap, int size, int align2, int startSearch) struct mem_block * -drmmmFindBlock(struct mem_block *heap, int start) +mmFindBlock(struct mem_block *heap, int start) { struct mem_block *p; @@ -235,7 +235,7 @@ Join2Blocks(struct mem_block *p) } int -drmmmFreeMem(struct mem_block *b) +mmFreeMem(struct mem_block *b) { if (!b) return 0; @@ -264,7 +264,7 @@ drmmmFreeMem(struct mem_block *b) void -drmmmDestroy(struct mem_block *heap) +mmDestroy(struct mem_block *heap) { struct mem_block *p; diff --git a/libdrm/intel/mm.h b/libdrm/intel/mm.h index 965bb0cd..49e3eecc 100644 --- a/libdrm/intel/mm.h +++ b/libdrm/intel/mm.h @@ -40,13 +40,21 @@ struct mem_block { unsigned int reserved:1; }; - +/* Rename the variables in the drm copy of this code so that it doesn't + * conflict with mesa or whoever else has copied it around. + */ +#define mmInit drm_mmInit +#define mmAllocMem drm_mmAllocMem +#define mmFreeMem drm_mmFreeMem +#define mmFindBlock drm_mmFindBlock +#define mmDestroy drm_mmDestroy +#define mmDumpMemInfo drm_mmDumpMemInfo /** * input: total size in bytes * return: a heap pointer if OK, NULL if error */ -extern struct mem_block *drmmmInit(int ofs, int size); +extern struct mem_block *mmInit(int ofs, int size); /** * Allocate 'size' bytes with 2^align2 bytes alignment, @@ -58,7 +66,7 @@ extern struct mem_block *drmmmInit(int ofs, int size); * startSearch = linear offset from start of heap to begin search * return: pointer to the allocated block, 0 if error */ -extern struct mem_block *drmmmAllocMem(struct mem_block *heap, int size, +extern struct mem_block *mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch); /** @@ -66,23 +74,23 @@ extern struct mem_block *drmmmAllocMem(struct mem_block *heap, int size, * input: pointer to a block * return: 0 if OK, -1 if error */ -extern int drmmmFreeMem(struct mem_block *b); +extern int mmFreeMem(struct mem_block *b); /** * Free block starts at offset * input: pointer to a heap, start offset * return: pointer to a block */ -extern struct mem_block *drmmmFindBlock(struct mem_block *heap, int start); +extern struct mem_block *mmFindBlock(struct mem_block *heap, int start); /** * destroy MM */ -extern void drmmmDestroy(struct mem_block *mmInit); +extern void mmDestroy(struct mem_block *mmInit); /** * For debuging purpose. */ -extern void drmmmDumpMemInfo(const struct mem_block *mmInit); +extern void mmDumpMemInfo(const struct mem_block *mmInit); #endif -- cgit v1.2.3 From b2606e325ac02782297def5ce27028c7fe2287c8 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 11 Jun 2008 16:19:23 -0700 Subject: [gem] Remove the drm_client_lock_take in set_domain. We no longer need to use it to protect against shared ringbuffer access. --- linux-core/drmP.h | 3 --- linux-core/drm_lock.c | 56 --------------------------------------------------- linux-core/i915_gem.c | 7 ++----- 3 files changed, 2 insertions(+), 64 deletions(-) diff --git a/linux-core/drmP.h b/linux-core/drmP.h index 48ff4b87..02826c8d 100644 --- a/linux-core/drmP.h +++ b/linux-core/drmP.h @@ -1169,9 +1169,6 @@ extern int drm_lock_take(struct drm_lock_data *lock_data, unsigned int context); extern int drm_lock_free(struct drm_lock_data *lock_data, unsigned int context); extern void drm_idlelock_take(struct drm_lock_data *lock_data); extern void drm_idlelock_release(struct drm_lock_data *lock_data); -extern int drm_client_lock_take(struct drm_device *dev, - struct drm_file *file_priv); -extern void drm_client_lock_release(struct drm_device *dev); /* * These are exported to drivers so that they can implement fencing using diff --git a/linux-core/drm_lock.c b/linux-core/drm_lock.c index e6eae42b..e8496a25 100644 --- a/linux-core/drm_lock.c +++ b/linux-core/drm_lock.c @@ -375,62 +375,6 @@ void drm_idlelock_release(struct drm_lock_data *lock_data) } EXPORT_SYMBOL(drm_idlelock_release); -/** - * Takes the lock on behalf of the client if needed, using the kernel context. - * - * This allows us to hide the hardware lock when it's required for protection - * of data structures (such as command ringbuffer) shared with the X Server, and - * a way for us to transition to lockless for those requests when the X Server - * stops accessing the ringbuffer directly, without having to update the - * other userland clients. - */ -int drm_client_lock_take(struct drm_device *dev, struct drm_file *file_priv) -{ - int ret; - unsigned long irqflags; - - /* If the client has the lock, we're already done. */ - if (drm_i_have_hw_lock(dev, file_priv)) - return 0; - - mutex_unlock (&dev->struct_mutex); - /* Client doesn't hold the lock. Block taking the lock with the kernel - * context on behalf of the client, and return whether we were - * successful. - */ - spin_lock_irqsave(&dev->lock.spinlock, irqflags); - dev->lock.user_waiters++; - spin_unlock_irqrestore(&dev->lock.spinlock, irqflags); - ret = wait_event_interruptible(dev->lock.lock_queue, - drm_lock_take(&dev->lock, - DRM_KERNEL_CONTEXT)); - spin_lock_irqsave(&dev->lock.spinlock, irqflags); - dev->lock.user_waiters--; - if (ret != 0) { - spin_unlock_irqrestore(&dev->lock.spinlock, irqflags); - } else { - dev->lock.file_priv = file_priv; - dev->lock.lock_time = jiffies; - dev->lock.kernel_held = 1; - file_priv->lock_count++; - spin_unlock_irqrestore(&dev->lock.spinlock, irqflags); - } - mutex_lock (&dev->struct_mutex); - return ret; -} -EXPORT_SYMBOL(drm_client_lock_take); - -void drm_client_lock_release(struct drm_device *dev) -{ - if (dev->lock.kernel_held) { - dev->lock.kernel_held = 0; - dev->lock.file_priv = NULL; - drm_lock_free(&dev->lock, DRM_KERNEL_CONTEXT); - } -} -EXPORT_SYMBOL(drm_client_lock_release); - - int drm_i_have_hw_lock(struct drm_device *dev, struct drm_file *file_priv) { diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index d608cb03..030ecf1d 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1930,14 +1930,11 @@ i915_gem_set_domain(struct drm_gem_object *obj, BUG_ON(!mutex_is_locked(&dev->struct_mutex)); - drm_client_lock_take(dev, file_priv); ret = i915_gem_object_set_domain(obj, read_domains, write_domain); - if (ret) { - drm_client_lock_release(dev); + if (ret) return ret; - } i915_gem_dev_set_domain(obj->dev); - drm_client_lock_release(dev); + return 0; } -- cgit v1.2.3 From cc74ac359053e95fd79d7011a2e18defcefb5c19 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Thu, 12 Jun 2008 10:56:37 -0700 Subject: Don't clear USER_INTERRUPT if we're not handling it. This fixes a regression from commit d434b64f6a760d85295e32298a9a1f3624ee1b69 which could cause us to fail to wake up for user interrupts if we lost a race. --- shared-core/i915_irq.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c index 0125f8b3..39da8143 100644 --- a/shared-core/i915_irq.c +++ b/shared-core/i915_irq.c @@ -507,7 +507,7 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) if (dev_priv->sarea_priv) dev_priv->sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv); - I915_WRITE(I915REG_INT_IDENTITY_R, iir | I915_USER_INTERRUPT); + I915_WRITE(I915REG_INT_IDENTITY_R, iir); (void) I915_READ(I915REG_INT_IDENTITY_R); /* Flush posted write */ if (iir & I915_USER_INTERRUPT) { -- cgit v1.2.3 From 100d6f3f83b559be000f69c04f9e5596cdc17c77 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Thu, 12 Jun 2008 13:22:40 -0700 Subject: Bump libdrm version to 2.4.0. We need a version to depend on from the other components that require GEM and the bufmgr code. Some interfaces will be removed before the 2.4.0 release. --- configure.ac | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/configure.ac b/configure.ac index a8855684..1cf877d5 100644 --- a/configure.ac +++ b/configure.ac @@ -19,7 +19,7 @@ # CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. AC_PREREQ(2.57) -AC_INIT([libdrm], 2.3.1, [dri-devel@lists.sourceforge.net], libdrm) +AC_INIT([libdrm], 2.4.0, [dri-devel@lists.sourceforge.net], libdrm) AC_CONFIG_SRCDIR([Makefile.am]) AM_INIT_AUTOMAKE([dist-bzip2]) -- cgit v1.2.3 From c892e26bdfcacfe7213085a08dd82e2cb7faa003 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Fri, 13 Jun 2008 09:49:05 -0700 Subject: [gem] Don't require the lock in execbuf now that it's not needed for the ring. --- linux-core/i915_gem.c | 2 -- 1 file changed, 2 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 030ecf1d..634b4f6c 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1595,8 +1595,6 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, uint64_t exec_offset; uint32_t seqno, flush_domains; - LOCK_TEST_WITH_RETURN(dev, file_priv); - #if WATCH_EXEC DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n", (int) args->buffers_ptr, args->buffer_count, args->batch_len); -- cgit v1.2.3 From 0558d99c01c38e24e97b57ef55f66bb94a92e157 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Fri, 13 Jun 2008 09:53:36 -0700 Subject: [gem] Remove the interrupt handler for retiring requests. This was insufficient once we started masking interrupts to only when someone was waiting for them (and would thus retire requests themselves). It was replaced by the retire_timer. --- shared-core/i915_dma.c | 2 -- shared-core/i915_drv.h | 3 --- shared-core/i915_irq.c | 23 ----------------------- 3 files changed, 28 deletions(-) diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c index 7e4de18b..667a6ac0 100644 --- a/shared-core/i915_dma.c +++ b/shared-core/i915_dma.c @@ -1077,8 +1077,6 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) init_timer_deferrable (&dev_priv->mm.retire_timer); INIT_WORK(&dev_priv->mm.retire_task, i915_gem_retire_handler); - INIT_WORK(&dev_priv->user_interrupt_task, - i915_user_interrupt_handler); dev_priv->mm.next_gem_seqno = 1; #ifdef __linux__ diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h index 9c9925b5..5af9a964 100644 --- a/shared-core/i915_drv.h +++ b/shared-core/i915_drv.h @@ -300,8 +300,6 @@ typedef struct drm_i915_private { */ int suspended; } mm; - - struct work_struct user_interrupt_task; } drm_i915_private_t; enum intel_chip_family { @@ -421,7 +419,6 @@ extern int i915_vblank_swap(struct drm_device *dev, void *data, struct drm_file *file_priv); extern void i915_user_irq_on(drm_i915_private_t *dev_priv); extern void i915_user_irq_off(drm_i915_private_t *dev_priv); -extern void i915_user_interrupt_handler(struct work_struct *work); /* i915_mem.c */ extern int i915_mem_alloc(struct drm_device *dev, void *data, diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c index 39da8143..6daf2918 100644 --- a/shared-core/i915_irq.c +++ b/shared-core/i915_irq.c @@ -435,28 +435,6 @@ u32 i915_get_vblank_counter(struct drm_device *dev, int plane) return count; } -/** - * Handler for user interrupts in process context (able to sleep, do VFS - * operations, etc. - * - * If another IRQ comes in while we're in this handler, it will still get put - * on the queue again to be rerun when we finish. - */ -void -i915_user_interrupt_handler(struct work_struct *work) -{ - drm_i915_private_t *dev_priv; - struct drm_device *dev; - - dev_priv = container_of(work, drm_i915_private_t, - user_interrupt_task); - dev = dev_priv->dev; - - mutex_lock(&dev->struct_mutex); - i915_gem_retire_requests(dev); - mutex_unlock(&dev->struct_mutex); -} - irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) { struct drm_device *dev = (struct drm_device *) arg; @@ -514,7 +492,6 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) DRM_WAKEUP(&dev_priv->irq_queue); #ifdef I915_HAVE_FENCE i915_fence_handler(dev); - schedule_work(&dev_priv->user_interrupt_task); #endif } -- cgit v1.2.3 From e558e1d7dacafa1e7f9681f1eaec072d663287de Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Fri, 13 Jun 2008 10:04:14 -0700 Subject: [gem] Catch -EINTR from blocking ioctls and restart them. Thanks to Thomas Hellstrom for catching the issue, no thanks to the kernel developer who authoritatively told me that they would get restarted on their own. --- libdrm/intel/intel_bufmgr_gem.c | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/libdrm/intel/intel_bufmgr_gem.c b/libdrm/intel/intel_bufmgr_gem.c index e057d949..5a28bd14 100644 --- a/libdrm/intel/intel_bufmgr_gem.c +++ b/libdrm/intel/intel_bufmgr_gem.c @@ -493,7 +493,10 @@ dri_gem_bo_map(dri_bo *bo, int write_enable) set_domain.handle = bo_gem->gem_handle; set_domain.read_domains = I915_GEM_DOMAIN_CPU; set_domain.write_domain = write_enable ? I915_GEM_DOMAIN_CPU : 0; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, &set_domain); + do { + ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN, + &set_domain); + } while (ret == -1 && errno == EINTR); if (ret != 0) { fprintf (stderr, "%s:%d: Error setting memory domains %d (%08x %08x): %s .\n", __FILE__, __LINE__, @@ -533,7 +536,9 @@ dri_gem_bo_subdata (dri_bo *bo, unsigned long offset, pwrite.offset = offset; pwrite.size = size; pwrite.data_ptr = (uint64_t) (uintptr_t) data; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PWRITE, &pwrite); + do { + ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PWRITE, &pwrite); + } while (ret == -1 && errno == EINTR); if (ret != 0) { fprintf (stderr, "%s:%d: Error writing data to buffer %d: (%d %d) %s .\n", __FILE__, __LINE__, @@ -557,7 +562,9 @@ dri_gem_bo_get_subdata (dri_bo *bo, unsigned long offset, pread.offset = offset; pread.size = size; pread.data_ptr = (uint64_t) (uintptr_t) data; - ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PREAD, &pread); + do { + ret = ioctl (bufmgr_gem->fd, DRM_IOCTL_I915_GEM_PREAD, &pread); + } while (ret == -1 && errno == EINTR); if (ret != 0) { fprintf (stderr, "%s:%d: Error reading data from buffer %d: (%d %d) %s .\n", __FILE__, __LINE__, -- cgit v1.2.3 From df4ef348c8a48cead807a42fcd315f7e422aa156 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Fri, 13 Jun 2008 10:47:33 -0700 Subject: Fix i915_wait_irq in the presence of interrupt masking. In the short-circuit code for the breadcrumb already being new enough, we need to update the sarea_priv copy of the breadcrumb just as if we had waited. Otherwise userland error checking will notice that we returned too early based on its wrong information, and call wait_irq again (leading to spinning until someone else comes along and updates the sarea_priv). This bug was hidden when we had interrupt masking disabled, such as in master, since the interrupt handler would update sarea_priv. --- shared-core/i915_irq.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c index 6daf2918..cfa9320e 100644 --- a/shared-core/i915_irq.c +++ b/shared-core/i915_irq.c @@ -560,8 +560,12 @@ int i915_wait_irq(struct drm_device * dev, int irq_nr) DRM_DEBUG("irq_nr=%d breadcrumb=%d\n", irq_nr, READ_BREADCRUMB(dev_priv)); - if (READ_BREADCRUMB(dev_priv) >= irq_nr) + if (READ_BREADCRUMB(dev_priv) >= irq_nr) { + if (dev_priv->sarea_priv) + dev_priv->sarea_priv->last_dispatch = + READ_BREADCRUMB(dev_priv); return 0; + } i915_user_irq_on(dev_priv); DRM_WAIT_ON(ret, dev_priv->irq_queue, 3 * DRM_HZ, -- cgit v1.2.3 From f378319b5627d84ca821c8e6a9f2fee612119477 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Tue, 10 Jun 2008 17:53:32 -0700 Subject: Use /bin/pwd instead of trusting shell built-in --- linux-core/Makefile | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/linux-core/Makefile b/linux-core/Makefile index 3af6f370..55dfb77c 100644 --- a/linux-core/Makefile +++ b/linux-core/Makefile @@ -116,7 +116,7 @@ V := $(shell if [ -f $(BOOTVERSION_PREFIX)version.h ]; then \ ifeq ($(V),"$(RUNNING_REL)") HEADERFROMBOOT := 1 -GETCONFIG := MAKEFILES=$(shell pwd)/.config +GETCONFIG := MAKEFILES=$(shell /bin/pwd)/.config HAVECONFIG := y endif @@ -163,7 +163,7 @@ endif all: modules modules: includes - +make -C $(LINUXDIR) $(GETCONFIG) SUBDIRS=`pwd` DRMSRCDIR=`pwd` modules + +make -C $(LINUXDIR) $(GETCONFIG) SUBDIRS=`/bin/pwd` DRMSRCDIR=`/bin/pwd` modules ifeq ($(HEADERFROMBOOT),1) @@ -239,7 +239,7 @@ drmstat: drmstat.c $(CC) $(PRGCFLAGS) $< -o $@ $(DRMSTATLIBS) install: - make -C $(LINUXDIR) $(GETCONFIG) SUBDIRS=`pwd` DRMSRCDIR=`pwd` modules_install + make -C $(LINUXDIR) $(GETCONFIG) SUBDIRS=`/bin/pwd` DRMSRCDIR=`/bin/pwd` modules_install else -- cgit v1.2.3 From e5364914ac2b785f9d806c72fff8d2ae914cad61 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Tue, 10 Jun 2008 18:11:15 -0700 Subject: [intel-gem] Reorder i915_add_request to schedule work last i915_add_request was calling schedule_delayed_work before adding the request to the list; it makes more sense to do that last. --- linux-core/i915_gem.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 634b4f6c..a14781f6 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -333,6 +333,7 @@ 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); @@ -360,11 +361,11 @@ i915_add_request(struct drm_device *dev, uint32_t flush_domains) request->seqno = seqno; request->emitted_jiffies = jiffies; request->flush_domains = flush_domains; - if (list_empty(&dev_priv->mm.request_list)) - mod_timer(&dev_priv->mm.retire_timer, jiffies + HZ); - + 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; } -- cgit v1.2.3 From 462af73149b9286a74b95b9cda5e4224ebe0dd87 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 09:19:30 -0700 Subject: [intel-gem] Use a delayed_work instead of a timer + work_struct We want request retirement to occur about once a second when the request queue is non-empty. This was done with a timer that queued a work_struct, using a delayed_work instead makes a lot more sense. --- linux-core/i915_gem.c | 15 +++------------ shared-core/i915_dma.c | 7 ++----- shared-core/i915_drv.h | 9 +++++---- 3 files changed, 10 insertions(+), 21 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index a14781f6..115f8e75 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -503,28 +503,19 @@ i915_gem_retire_requests(struct drm_device *dev) } void -i915_gem_retire_timeout(unsigned long data) -{ - struct drm_device *dev = (struct drm_device *) data; - drm_i915_private_t *dev_priv = dev->dev_private; - - schedule_work(&dev_priv->mm.retire_task); -} - -void -i915_gem_retire_handler(struct work_struct *work) +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_task); + 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)) - mod_timer(&dev_priv->mm.retire_timer, jiffies + HZ); + schedule_delayed_work (&dev_priv->mm.retire_work, HZ); mutex_unlock(&dev->struct_mutex); } diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c index 667a6ac0..669f1e4e 100644 --- a/shared-core/i915_dma.c +++ b/shared-core/i915_dma.c @@ -1072,11 +1072,8 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) INIT_LIST_HEAD(&dev_priv->mm.flushing_list); INIT_LIST_HEAD(&dev_priv->mm.inactive_list); INIT_LIST_HEAD(&dev_priv->mm.request_list); - dev_priv->mm.retire_timer.function = i915_gem_retire_timeout; - dev_priv->mm.retire_timer.data = (unsigned long) dev; - init_timer_deferrable (&dev_priv->mm.retire_timer); - INIT_WORK(&dev_priv->mm.retire_task, - i915_gem_retire_handler); + INIT_DELAYED_WORK(&dev_priv->mm.retire_work, + i915_gem_retire_work_handler); dev_priv->mm.next_gem_seqno = 1; #ifdef __linux__ diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h index 5af9a964..5cc6e680 100644 --- a/shared-core/i915_drv.h +++ b/shared-core/i915_drv.h @@ -285,8 +285,7 @@ typedef struct drm_i915_private { * fire periodically while the ring is running. When it * fires, go retire requests. */ - struct timer_list retire_timer; - struct work_struct retire_task; + struct delayed_work retire_work; uint32_t next_gem_seqno; @@ -488,8 +487,7 @@ 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); void i915_gem_retire_requests(struct drm_device *dev); -void i915_gem_retire_timeout(unsigned long data); -void i915_gem_retire_handler(struct work_struct *work); +void i915_gem_retire_work_handler(struct work_struct *work); #endif #ifdef __linux__ @@ -915,6 +913,9 @@ extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller); #define XY_SRC_COPY_BLT_SRC_TILED (1<<15) #define XY_SRC_COPY_BLT_DST_TILED (1<<11) +#define MI_NOOP (0) +#define MI_NOOP_ENABLE_NOPID (1 << 22) +#define MI_NOOP_ID_MASK ((1 << 22) - 1) #define MI_BATCH_BUFFER ((0x30<<23)|1) #define MI_BATCH_BUFFER_START (0x31<<23) -- cgit v1.2.3 From 5957470ca3be6c0225985f74b1511401e02c014b Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Tue, 10 Jun 2008 18:19:19 -0700 Subject: [intel] Restructure irq to pend all work until after iir write. The interrupt identity register must be writen before any work occurs lest we drop an interrupt on the floor. This patch just shuffles code around to make sure that IIR is written as early as possible. --- shared-core/i915_irq.c | 46 ++++++++++++++++++++++++---------------------- 1 file changed, 24 insertions(+), 22 deletions(-) diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c index cfa9320e..58781a4a 100644 --- a/shared-core/i915_irq.c +++ b/shared-core/i915_irq.c @@ -440,7 +440,7 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) struct drm_device *dev = (struct drm_device *) arg; drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; u32 iir; - u32 pipea_stats, pipeb_stats; + u32 pipea_stats = 0, pipeb_stats = 0; int vblank = 0; iir = I915_READ(I915REG_INT_IDENTITY_R); @@ -463,31 +463,19 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) */ if (iir & I915_DISPLAY_PIPE_A_EVENT_INTERRUPT) { pipea_stats = I915_READ(I915REG_PIPEASTAT); - if (pipea_stats & (I915_START_VBLANK_INTERRUPT_STATUS| - I915_VBLANK_INTERRUPT_STATUS)) - { - vblank++; - drm_handle_vblank(dev, i915_get_plane(dev, 0)); - } I915_WRITE(I915REG_PIPEASTAT, pipea_stats); } if (iir & I915_DISPLAY_PIPE_B_EVENT_INTERRUPT) { pipeb_stats = I915_READ(I915REG_PIPEBSTAT); - if (pipeb_stats & (I915_START_VBLANK_INTERRUPT_STATUS| - I915_VBLANK_INTERRUPT_STATUS)) - { - vblank++; - drm_handle_vblank(dev, i915_get_plane(dev, 1)); - } I915_WRITE(I915REG_PIPEBSTAT, pipeb_stats); } + I915_WRITE(I915REG_INT_IDENTITY_R, iir); + (void) I915_READ(I915REG_INT_IDENTITY_R); /* Flush posted writes */ + if (dev_priv->sarea_priv) dev_priv->sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv); - I915_WRITE(I915REG_INT_IDENTITY_R, iir); - (void) I915_READ(I915REG_INT_IDENTITY_R); /* Flush posted write */ - if (iir & I915_USER_INTERRUPT) { DRM_WAKEUP(&dev_priv->irq_queue); #ifdef I915_HAVE_FENCE @@ -495,6 +483,16 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) #endif } + if (pipea_stats & (I915_START_VBLANK_INTERRUPT_STATUS| + I915_VBLANK_INTERRUPT_STATUS)) { + vblank = 1; + drm_handle_vblank(dev, i915_get_plane(dev, 0)); + } + if (pipeb_stats & (I915_START_VBLANK_INTERRUPT_STATUS| + I915_VBLANK_INTERRUPT_STATUS)) { + vblank = 1; + drm_handle_vblank(dev, i915_get_plane(dev, 1)); + } if (vblank) { if (dev_priv->swaps_pending > 0) drm_locked_tasklet(dev, i915_vblank_tasklet); @@ -526,9 +524,11 @@ void i915_user_irq_on(drm_i915_private_t *dev_priv) { DRM_SPINLOCK(&dev_priv->user_irq_lock); if (dev_priv->irq_enabled && (++dev_priv->user_irq_refcount == 1)){ - dev_priv->irq_mask_reg &= ~I915_USER_INTERRUPT; - I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); - (void) I915_READ (I915REG_INT_ENABLE_R); + if ((dev_priv->irq_mask_reg & I915_USER_INTERRUPT) != 0) { + dev_priv->irq_mask_reg &= ~I915_USER_INTERRUPT; + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + (void) I915_READ (I915REG_INT_MASK_R); + } } DRM_SPINUNLOCK(&dev_priv->user_irq_lock); @@ -539,9 +539,11 @@ void i915_user_irq_off(drm_i915_private_t *dev_priv) DRM_SPINLOCK(&dev_priv->user_irq_lock); BUG_ON(dev_priv->irq_enabled && dev_priv->user_irq_refcount <= 0); if (dev_priv->irq_enabled && (--dev_priv->user_irq_refcount == 0)) { - dev_priv->irq_mask_reg |= I915_USER_INTERRUPT; - I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); - (void) I915_READ(I915REG_INT_MASK_R); + if ((dev_priv->irq_mask_reg & I915_USER_INTERRUPT) == 0) { + dev_priv->irq_mask_reg |= I915_USER_INTERRUPT; + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + (void) I915_READ(I915REG_INT_MASK_R); + } } DRM_SPINUNLOCK(&dev_priv->user_irq_lock); } -- cgit v1.2.3 From 3762c9ea6754763694b1de4df3acd9dc37247f87 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 10:27:39 -0700 Subject: [intel] Enable MSI for i915 IRQ --- linux-core/i915_drv.c | 17 +++++++++++++++-- shared-core/i915_irq.c | 20 ++++++++++++++++---- 2 files changed, 31 insertions(+), 6 deletions(-) diff --git a/linux-core/i915_drv.c b/linux-core/i915_drv.c index 8f51cd4d..89f089b0 100644 --- a/linux-core/i915_drv.c +++ b/linux-core/i915_drv.c @@ -560,6 +560,7 @@ static int i915_resume(struct drm_device *dev) } static int probe(struct pci_dev *pdev, const struct pci_device_id *ent); +static void remove(struct pci_dev *pdev); static struct drm_driver driver = { /* don't use mtrr's here, the Xserver or user space app should * deal with them for intel hardware. @@ -604,7 +605,7 @@ static struct drm_driver driver = { .name = DRIVER_NAME, .id_table = pciidlist, .probe = probe, - .remove = __devexit_p(drm_cleanup_pci), + .remove = remove, }, #ifdef I915_HAVE_FENCE .fence_driver = &i915_fence_driver, @@ -622,7 +623,19 @@ static struct drm_driver driver = { static int probe(struct pci_dev *pdev, const struct pci_device_id *ent) { - return drm_get_dev(pdev, ent, &driver); + int ret; + + (void) pci_enable_msi(pdev); + ret = drm_get_dev(pdev, ent, &driver); + if (ret && pdev->msi_enabled) + pci_disable_msi(pdev); + return ret; +} +static void remove(struct pci_dev *pdev) +{ + if (pdev->msi_enabled) + pci_disable_msi(pdev); + drm_cleanup_pci(pdev); } static int __init i915_init(void) diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c index 58781a4a..9ba5b00a 100644 --- a/shared-core/i915_irq.c +++ b/shared-core/i915_irq.c @@ -33,6 +33,13 @@ #define MAX_NOPID ((u32)~0) +/* + * These are the interrupts used by the driver + */ +#define I915_INTERRUPT_ENABLE_MASK (I915_USER_INTERRUPT | \ + I915_DISPLAY_PIPE_A_EVENT_INTERRUPT | \ + I915_DISPLAY_PIPE_B_EVENT_INTERRUPT) + /** * i915_get_pipe - return the the pipe associated with a given plane * @dev: DRM device @@ -443,6 +450,8 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) u32 pipea_stats = 0, pipeb_stats = 0; int vblank = 0; + if (dev->pdev->msi_enabled) + I915_WRITE(I915REG_INT_ENABLE_R, 0); iir = I915_READ(I915REG_INT_IDENTITY_R); #if 0 DRM_DEBUG("flag=%08x\n", iir); @@ -454,6 +463,9 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) I915_READ(I915REG_INT_ENABLE_R), I915_READ(I915REG_PIPEASTAT), I915_READ(I915REG_PIPEBSTAT)); + if (dev->pdev->msi_enabled) + I915_WRITE(I915REG_INT_ENABLE_R, + I915_INTERRUPT_ENABLE_MASK); return IRQ_NONE; } @@ -498,6 +510,8 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) drm_locked_tasklet(dev, i915_vblank_tasklet); } + if (dev->pdev->msi_enabled) + I915_WRITE(I915REG_INT_ENABLE_R, I915_INTERRUPT_ENABLE_MASK); return IRQ_HANDLED; } @@ -724,11 +738,9 @@ static void i915_enable_interrupt (struct drm_device *dev) { drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; - dev_priv->irq_mask_reg = (I915_USER_INTERRUPT | - I915_DISPLAY_PIPE_A_EVENT_INTERRUPT | - I915_DISPLAY_PIPE_B_EVENT_INTERRUPT); + dev_priv->irq_mask_reg = I915_INTERRUPT_ENABLE_MASK; I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); - I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_mask_reg); + I915_WRITE(I915REG_INT_ENABLE_R, I915_INTERRUPT_ENABLE_MASK); (void) I915_READ (I915REG_INT_ENABLE_R); dev_priv->irq_enabled = 1; } -- cgit v1.2.3 From 6b2cba1ecc5f9f289b5d91e229b7f7b0999bee5b Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 11:33:27 -0700 Subject: [intel-gem] evict_something was failing when wait_request freed objects When i915_wait_request clears object from the active list, it may end up freeing them and not moving them to the inactive list. This ends up unbinding objects from the GTT without there ever being new objects visible to i915_gem_evict_something on the inactive list. As the only success condition required the presence of objects on the inactive list, this would falsely assume that no GTT space had been made available, and end up returning -ENOMEM to the application. --- linux-core/i915_gem.c | 39 ++++++++++++++++++++++----------------- 1 file changed, 22 insertions(+), 17 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 115f8e75..b114192d 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -805,7 +805,7 @@ 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; + int ret = 0; for (;;) { /* If there's an inactive buffer available now, grab it @@ -817,6 +817,13 @@ i915_gem_evict_something(struct drm_device *dev) 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; } @@ -826,17 +833,21 @@ i915_gem_evict_something(struct drm_device *dev) */ if (!list_empty(&dev_priv->mm.request_list)) { struct drm_i915_gem_request *request; - int ret; request = list_first_entry(&dev_priv->mm.request_list, struct drm_i915_gem_request, list); ret = i915_wait_request(dev, request->seqno); - if (ret != 0) - return ret; - - continue; + + /* 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 @@ -859,21 +870,15 @@ i915_gem_evict_something(struct drm_device *dev) 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; } - -#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); - return ret; } @@ -959,7 +964,7 @@ i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment) ret = i915_gem_evict_something(dev); if (ret != 0) { - DRM_ERROR("Failed to evict a buffer\n"); + DRM_ERROR("Failed to evict a buffer %d\n", ret); return ret; } goto search_free; -- cgit v1.2.3 From ced9ebf64543b4d64a38feee3293040af953acc0 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 12:06:13 -0700 Subject: [intel-gem] throttle based on frames rather than time. Reduces jitter. Record the last execbuffer sequence for each client. Record that sequence in the throttle ioctl as the 'throttle sequence'. Wait for the last throttle sequence in the throttle ioctl. --- linux-core/i915_drv.c | 3 +++ linux-core/i915_gem.c | 34 ++++++++++------------------------ shared-core/i915_dma.c | 26 ++++++++++++++++++++++++++ shared-core/i915_drv.h | 10 ++++++++++ 4 files changed, 49 insertions(+), 24 deletions(-) diff --git a/linux-core/i915_drv.c b/linux-core/i915_drv.c index 89f089b0..012ca822 100644 --- a/linux-core/i915_drv.c +++ b/linux-core/i915_drv.c @@ -561,6 +561,7 @@ static int i915_resume(struct drm_device *dev) static int probe(struct pci_dev *pdev, const struct pci_device_id *ent); static void remove(struct pci_dev *pdev); + static struct drm_driver driver = { /* don't use mtrr's here, the Xserver or user space app should * deal with them for intel hardware. @@ -571,8 +572,10 @@ static struct drm_driver driver = { .load = i915_driver_load, .unload = i915_driver_unload, .firstopen = i915_driver_firstopen, + .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, diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index b114192d..c2d1fab6 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1548,33 +1548,17 @@ i915_dispatch_gem_execbuffer(struct drm_device *dev, * relatively low latency when blocking on a particular request to finish. */ static int -i915_gem_ring_throttle(struct drm_device *dev) +i915_gem_ring_throttle(struct drm_device *dev, 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; int ret = 0; + uint32_t seqno; mutex_lock(&dev->struct_mutex); - while (!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); - - /* Break out if we're close enough. */ - if ((long) (jiffies - request->emitted_jiffies) <= - (20 * HZ) / 1000) { - mutex_unlock(&dev->struct_mutex); - return 0; - } - - /* Wait on the last request if not. */ - ret = i915_wait_request(dev, request->seqno); - if (ret != 0) { - mutex_unlock(&dev->struct_mutex); - return ret; - } - } + 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; } @@ -1584,6 +1568,7 @@ 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; @@ -1724,6 +1709,7 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, */ 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; @@ -1881,7 +1867,7 @@ int i915_gem_throttle_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { - return i915_gem_ring_throttle(dev); + return i915_gem_ring_throttle(dev, file_priv); } int i915_gem_init_object(struct drm_gem_object *obj) diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c index 669f1e4e..2157c594 100644 --- a/shared-core/i915_dma.c +++ b/shared-core/i915_dma.c @@ -1149,12 +1149,38 @@ void i915_driver_lastclose(struct drm_device * dev) i915_dma_cleanup(dev); } +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_preclose(struct drm_device * dev, struct drm_file *file_priv) { drm_i915_private_t *dev_priv = dev->dev_private; 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), diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h index 5cc6e680..334bc43f 100644 --- a/shared-core/i915_drv.h +++ b/shared-core/i915_drv.h @@ -301,6 +301,13 @@ typedef struct drm_i915_private { } mm; } drm_i915_private_t; +struct drm_i915_file_private { + struct { + uint32_t last_gem_seqno; + uint32_t last_gem_throttle_seqno; + } mm; +}; + enum intel_chip_family { CHIP_I8XX = 0x01, CHIP_I9XX = 0x02, @@ -378,8 +385,11 @@ 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 void i915_driver_lastclose(struct drm_device * dev); +extern int i915_driver_open(struct drm_device *dev, struct drm_file *file_priv); 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); -- cgit v1.2.3 From baf521369478eff2842b99feda16f9d145402d27 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 14:28:18 -0700 Subject: [intel-gem] Pin objects during execbuffer Pinning the objects avoids accidentally evicting them while binding other objects. --- linux-core/i915_gem.c | 76 +++++++++++++++++++++++++++++++++++---------------- 1 file changed, 52 insertions(+), 24 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index c2d1fab6..d14b2f2d 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1298,27 +1298,25 @@ i915_gem_object_check_coherency(struct drm_gem_object *obj, int handle) #endif /** - * Bind an object to the GTT and evaluate the relocations landing in it. + * Pin an object to the GTT and evaluate the relocations landing in it. */ static int -i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, - struct drm_file *file_priv, - struct drm_i915_gem_exec_object *entry) +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; + int i, ret; uint32_t last_reloc_offset = -1; void *reloc_page = NULL; /* Choose the GTT offset for our buffer and put it there. */ - if (obj_priv->gtt_space == NULL) { - i915_gem_object_bind_to_gtt(obj, (unsigned) entry->alignment); - if (obj_priv->gtt_space == NULL) - return -ENOMEM; - } + ret = i915_gem_object_pin(obj, (uint32_t) entry->alignment); + if (ret) + return ret; entry->offset = obj_priv->gtt_offset; @@ -1334,13 +1332,17 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, int ret; ret = copy_from_user(&reloc, relocs + i, sizeof(reloc)); - if (ret != 0) + 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) + if (target_obj == NULL) { + i915_gem_object_unpin(obj); return -EINVAL; + } target_obj_priv = target_obj->driver_private; /* The target buffer should have appeared before us in the @@ -1350,6 +1352,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, 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; } @@ -1359,6 +1362,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, 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) { @@ -1367,6 +1371,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, obj, reloc.target_handle, (int) reloc.offset); drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); return -EINVAL; } @@ -1380,6 +1385,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, reloc.write_domain, target_obj->pending_write_domain); drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); return -EINVAL; } @@ -1440,6 +1446,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, last_reloc_offset = reloc_offset; if (reloc_page == NULL) { drm_gem_object_unreference(target_obj); + i915_gem_object_unpin(obj); return -ENOMEM; } } @@ -1462,6 +1469,7 @@ i915_gem_object_bind_and_relocate(struct drm_gem_object *obj, 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; } @@ -1573,7 +1581,7 @@ i915_gem_execbuffer(struct drm_device *dev, void *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; + int ret, i, pinned = 0; uint64_t exec_offset; uint32_t seqno, flush_domains; @@ -1632,13 +1640,14 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, object_list[i]->pending_read_domains = 0; object_list[i]->pending_write_domain = 0; - ret = i915_gem_object_bind_and_relocate(object_list[i], - file_priv, - &exec_list[i]); + 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; } /* Set the pending read domains for the batch buffer to COMMAND */ @@ -1735,6 +1744,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, 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]); } @@ -1752,32 +1763,47 @@ pre_mutex_err: 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; if (obj_priv->gtt_space == NULL) { ret = i915_gem_object_bind_to_gtt(obj, alignment); if (ret != 0) { - DRM_ERROR("Failure to bind in " - "i915_gem_pin_ioctl(): %d\n", - ret); - drm_gem_object_unreference(obj); - mutex_unlock(&dev->struct_mutex); + 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 && + !obj_priv->active && + obj->write_domain == 0) + list_del_init(&obj_priv->list); + 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; 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 && + !obj_priv->active && obj->write_domain == 0) + list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list); } int @@ -1963,8 +1989,10 @@ i915_gem_init_ringbuffer(struct drm_device *dev) obj_priv = obj->driver_private; ret = i915_gem_object_pin(obj, 4096); - if (ret != 0) + if (ret != 0) { + drm_gem_object_unreference(obj); return ret; + } /* Set up the kernel mapping for the ring. */ dev_priv->ring.Size = obj->size; -- cgit v1.2.3 From 4086cdb6550a4e957fd436c77a6260204e026538 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 15:38:13 -0700 Subject: [intel-gem] Left the last exec buffer pinned. oops. Loop end variable 'pinned' was set one too low. --- linux-core/i915_gem.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index d14b2f2d..210ae9d9 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1647,7 +1647,7 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, DRM_ERROR("object bind and relocate failed %d\n", ret); goto err; } - pinned = i; + pinned = i + 1; } /* Set the pending read domains for the batch buffer to COMMAND */ @@ -1745,7 +1745,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, err: if (object_list != NULL) { for (i = 0; i < pinned; i++) - i915_gem_object_unpin (object_list[i]); + i915_gem_object_unpin(object_list[i]); + for (i = 0; i < args->buffer_count; i++) drm_gem_object_unreference(object_list[i]); } -- cgit v1.2.3 From 217beb9c8de01417ac6219b54bd25046da6d4c7a Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 15:43:02 -0700 Subject: [intel-gem] add gtt and pin counts to /proc/dri/*/gem_objects Not quite portable, but these are useful for intel. Some more general mechanism could be done... --- linux-core/drmP.h | 5 +++++ linux-core/drm_gem.c | 7 +++++++ linux-core/drm_proc.c | 4 ++++ linux-core/i915_gem.c | 26 +++++++++++++++++++------- 4 files changed, 35 insertions(+), 7 deletions(-) diff --git a/linux-core/drmP.h b/linux-core/drmP.h index 02826c8d..ba25f3a8 100644 --- a/linux-core/drmP.h +++ b/linux-core/drmP.h @@ -965,6 +965,11 @@ struct drm_device { spinlock_t object_name_lock; struct idr object_name_idr; atomic_t object_count; + atomic_t object_memory; + atomic_t pin_count; + atomic_t pin_memory; + atomic_t gtt_count; + atomic_t gtt_memory; uint32_t invalidate_domains; /* domains pending invalidation */ uint32_t flush_domains; /* domains pending flush */ /*@} */ diff --git a/linux-core/drm_gem.c b/linux-core/drm_gem.c index 47b55434..434155b3 100644 --- a/linux-core/drm_gem.c +++ b/linux-core/drm_gem.c @@ -74,6 +74,11 @@ 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; } @@ -106,6 +111,7 @@ drm_gem_object_alloc(struct drm_device *dev, size_t size) return NULL; } atomic_inc(&dev->object_count); + atomic_add(obj->size, &dev->object_memory); return obj; } EXPORT_SYMBOL(drm_gem_object_alloc); @@ -376,6 +382,7 @@ drm_gem_object_free(struct kref *kref) fput(obj->filp); atomic_dec(&dev->object_count); + atomic_sub(obj->size, &dev->object_memory); kfree(obj); } EXPORT_SYMBOL(drm_gem_object_free); diff --git a/linux-core/drm_proc.c b/linux-core/drm_proc.c index b1b976b2..71d55274 100644 --- a/linux-core/drm_proc.c +++ b/linux-core/drm_proc.c @@ -655,6 +655,10 @@ static int drm_gem_object_info(char *buf, char **start, off_t offset, *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)); if (len > request + offset) return request; *eof = 1; diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 210ae9d9..cb5d663e 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -677,6 +677,7 @@ i915_gem_object_wait_rendering(struct drm_gem_object *obj) 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; @@ -706,6 +707,9 @@ i915_gem_object_unbind(struct drm_gem_object *obj) i915_gem_object_free_page_list(obj); + atomic_dec(&dev->gtt_count); + atomic_sub(obj->size, &dev->gtt_memory); + drm_memrange_put_block(obj_priv->gtt_space); obj_priv->gtt_space = NULL; @@ -995,6 +999,8 @@ i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment) 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 @@ -1764,6 +1770,7 @@ pre_mutex_err: 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; @@ -1779,10 +1786,12 @@ i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment) /* If the object is not active and not pending a flush, * remove it from the inactive list */ - if (obj_priv->pin_count == 1 && - !obj_priv->active && - obj->write_domain == 0) - list_del_init(&obj_priv->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 == 0) + list_del_init(&obj_priv->list); + } return 0; } @@ -1802,9 +1811,12 @@ i915_gem_object_unpin(struct drm_gem_object *obj) * neither active nor being flushed, then stick it on * the inactive list */ - if (obj_priv->pin_count == 0 && - !obj_priv->active && obj->write_domain == 0) - list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list); + if (obj_priv->pin_count == 0) { + if (!obj_priv->active && obj->write_domain == 0) + list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list); + atomic_dec(&dev->pin_count); + atomic_sub(obj->size, &dev->pin_memory); + } } int -- cgit v1.2.3 From 8b9ab108ec1f2ba2b503f713769c4946849b3cb2 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 16:03:22 -0700 Subject: [libdrm] Restart all ioctls on signal receipt Receiving a signal should be ignored by the library, so just restart any ioctl which returns EINTR or EAGAIN. --- libdrm/xf86drm.c | 172 +++++++++++++++++++++++++++++-------------------------- 1 file changed, 92 insertions(+), 80 deletions(-) diff --git a/libdrm/xf86drm.c b/libdrm/xf86drm.c index 0383b6b8..edb0c90b 100644 --- a/libdrm/xf86drm.c +++ b/libdrm/xf86drm.c @@ -171,6 +171,19 @@ static char *drmStrdup(const char *s) return retval; } +/** + * Call ioctl, restarting if it is interupted + */ +static int +drmIoctl(int fd, int request, void *arg) +{ + int ret; + + do { + ret = ioctl(fd, request, arg); + } while (ret == -1 && (errno == EINTR || errno == EAGAIN)); + return ret; +} static unsigned long drmGetKeyFromFd(int fd) { @@ -668,7 +681,7 @@ drmVersionPtr drmGetVersion(int fd) version->desc_len = 0; version->desc = NULL; - if (ioctl(fd, DRM_IOCTL_VERSION, version)) { + if (drmIoctl(fd, DRM_IOCTL_VERSION, version)) { drmFreeKernelVersion(version); return NULL; } @@ -680,7 +693,7 @@ drmVersionPtr drmGetVersion(int fd) if (version->desc_len) version->desc = drmMalloc(version->desc_len + 1); - if (ioctl(fd, DRM_IOCTL_VERSION, version)) { + if (drmIoctl(fd, DRM_IOCTL_VERSION, version)) { drmMsg("DRM_IOCTL_VERSION: %s\n", strerror(errno)); drmFreeKernelVersion(version); return NULL; @@ -766,10 +779,10 @@ char *drmGetBusid(int fd) u.unique_len = 0; u.unique = NULL; - if (ioctl(fd, DRM_IOCTL_GET_UNIQUE, &u)) + if (drmIoctl(fd, DRM_IOCTL_GET_UNIQUE, &u)) return NULL; u.unique = drmMalloc(u.unique_len + 1); - if (ioctl(fd, DRM_IOCTL_GET_UNIQUE, &u)) + if (drmIoctl(fd, DRM_IOCTL_GET_UNIQUE, &u)) return NULL; u.unique[u.unique_len] = '\0'; @@ -796,7 +809,7 @@ int drmSetBusid(int fd, const char *busid) u.unique = (char *)busid; u.unique_len = strlen(busid); - if (ioctl(fd, DRM_IOCTL_SET_UNIQUE, &u)) { + if (drmIoctl(fd, DRM_IOCTL_SET_UNIQUE, &u)) { return -errno; } return 0; @@ -807,7 +820,7 @@ int drmGetMagic(int fd, drm_magic_t * magic) drm_auth_t auth; *magic = 0; - if (ioctl(fd, DRM_IOCTL_GET_MAGIC, &auth)) + if (drmIoctl(fd, DRM_IOCTL_GET_MAGIC, &auth)) return -errno; *magic = auth.magic; return 0; @@ -818,7 +831,7 @@ int drmAuthMagic(int fd, drm_magic_t magic) drm_auth_t auth; auth.magic = magic; - if (ioctl(fd, DRM_IOCTL_AUTH_MAGIC, &auth)) + if (drmIoctl(fd, DRM_IOCTL_AUTH_MAGIC, &auth)) return -errno; return 0; } @@ -883,7 +896,7 @@ int drmAddMap(int fd, drm_handle_t offset, drmSize size, drmMapType type, map.handle = 0; map.type = type; map.flags = flags; - if (ioctl(fd, DRM_IOCTL_ADD_MAP, &map)) + if (drmIoctl(fd, DRM_IOCTL_ADD_MAP, &map)) return -errno; if (handle) *handle = (drm_handle_t)map.handle; @@ -896,7 +909,7 @@ int drmRmMap(int fd, drm_handle_t handle) map.handle = (void *)handle; - if(ioctl(fd, DRM_IOCTL_RM_MAP, &map)) + if(drmIoctl(fd, DRM_IOCTL_RM_MAP, &map)) return -errno; return 0; } @@ -929,7 +942,7 @@ int drmAddBufs(int fd, int count, int size, drmBufDescFlags flags, request.flags = flags; request.agp_start = agp_offset; - if (ioctl(fd, DRM_IOCTL_ADD_BUFS, &request)) + if (drmIoctl(fd, DRM_IOCTL_ADD_BUFS, &request)) return -errno; return request.count; } @@ -942,7 +955,7 @@ int drmMarkBufs(int fd, double low, double high) info.count = 0; info.list = NULL; - if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) + if (drmIoctl(fd, DRM_IOCTL_INFO_BUFS, &info)) return -EINVAL; if (!info.count) @@ -951,7 +964,7 @@ int drmMarkBufs(int fd, double low, double high) if (!(info.list = drmMalloc(info.count * sizeof(*info.list)))) return -ENOMEM; - if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) { + if (drmIoctl(fd, DRM_IOCTL_INFO_BUFS, &info)) { int retval = -errno; drmFree(info.list); return retval; @@ -960,7 +973,7 @@ int drmMarkBufs(int fd, double low, double high) for (i = 0; i < info.count; i++) { info.list[i].low_mark = low * info.list[i].count; info.list[i].high_mark = high * info.list[i].count; - if (ioctl(fd, DRM_IOCTL_MARK_BUFS, &info.list[i])) { + if (drmIoctl(fd, DRM_IOCTL_MARK_BUFS, &info.list[i])) { int retval = -errno; drmFree(info.list); return retval; @@ -992,7 +1005,7 @@ int drmFreeBufs(int fd, int count, int *list) request.count = count; request.list = list; - if (ioctl(fd, DRM_IOCTL_FREE_BUFS, &request)) + if (drmIoctl(fd, DRM_IOCTL_FREE_BUFS, &request)) return -errno; return 0; } @@ -1081,14 +1094,14 @@ drmBufInfoPtr drmGetBufInfo(int fd) info.count = 0; info.list = NULL; - if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) + if (drmIoctl(fd, DRM_IOCTL_INFO_BUFS, &info)) return NULL; if (info.count) { if (!(info.list = drmMalloc(info.count * sizeof(*info.list)))) return NULL; - if (ioctl(fd, DRM_IOCTL_INFO_BUFS, &info)) { + if (drmIoctl(fd, DRM_IOCTL_INFO_BUFS, &info)) { drmFree(info.list); return NULL; } @@ -1132,7 +1145,7 @@ drmBufMapPtr drmMapBufs(int fd) bufs.count = 0; bufs.list = NULL; bufs.virtual = NULL; - if (ioctl(fd, DRM_IOCTL_MAP_BUFS, &bufs)) + if (drmIoctl(fd, DRM_IOCTL_MAP_BUFS, &bufs)) return NULL; if (!bufs.count) @@ -1141,7 +1154,7 @@ drmBufMapPtr drmMapBufs(int fd) if (!(bufs.list = drmMalloc(bufs.count * sizeof(*bufs.list)))) return NULL; - if (ioctl(fd, DRM_IOCTL_MAP_BUFS, &bufs)) { + if (drmIoctl(fd, DRM_IOCTL_MAP_BUFS, &bufs)) { drmFree(bufs.list); return NULL; } @@ -1256,7 +1269,7 @@ int drmGetLock(int fd, drm_context_t context, drmLockFlags flags) if (flags & DRM_HALT_ALL_QUEUES) lock.flags |= _DRM_HALT_ALL_QUEUES; if (flags & DRM_HALT_CUR_QUEUES) lock.flags |= _DRM_HALT_CUR_QUEUES; - while (ioctl(fd, DRM_IOCTL_LOCK, &lock)) + while (drmIoctl(fd, DRM_IOCTL_LOCK, &lock)) ; return 0; } @@ -1279,7 +1292,7 @@ int drmUnlock(int fd, drm_context_t context) lock.context = context; lock.flags = 0; - return ioctl(fd, DRM_IOCTL_UNLOCK, &lock); + return drmIoctl(fd, DRM_IOCTL_UNLOCK, &lock); } drm_context_t *drmGetReservedContextList(int fd, int *count) @@ -1291,7 +1304,7 @@ drm_context_t *drmGetReservedContextList(int fd, int *count) res.count = 0; res.contexts = NULL; - if (ioctl(fd, DRM_IOCTL_RES_CTX, &res)) + if (drmIoctl(fd, DRM_IOCTL_RES_CTX, &res)) return NULL; if (!res.count) @@ -1305,7 +1318,7 @@ drm_context_t *drmGetReservedContextList(int fd, int *count) } res.contexts = list; - if (ioctl(fd, DRM_IOCTL_RES_CTX, &res)) + if (drmIoctl(fd, DRM_IOCTL_RES_CTX, &res)) return NULL; for (i = 0; i < res.count; i++) @@ -1344,7 +1357,7 @@ int drmCreateContext(int fd, drm_context_t *handle) drm_ctx_t ctx; ctx.flags = 0; /* Modified with functions below */ - if (ioctl(fd, DRM_IOCTL_ADD_CTX, &ctx)) + if (drmIoctl(fd, DRM_IOCTL_ADD_CTX, &ctx)) return -errno; *handle = ctx.handle; return 0; @@ -1355,7 +1368,7 @@ int drmSwitchToContext(int fd, drm_context_t context) drm_ctx_t ctx; ctx.handle = context; - if (ioctl(fd, DRM_IOCTL_SWITCH_CTX, &ctx)) + if (drmIoctl(fd, DRM_IOCTL_SWITCH_CTX, &ctx)) return -errno; return 0; } @@ -1376,7 +1389,7 @@ int drmSetContextFlags(int fd, drm_context_t context, drm_context_tFlags flags) ctx.flags |= _DRM_CONTEXT_PRESERVED; if (flags & DRM_CONTEXT_2DONLY) ctx.flags |= _DRM_CONTEXT_2DONLY; - if (ioctl(fd, DRM_IOCTL_MOD_CTX, &ctx)) + if (drmIoctl(fd, DRM_IOCTL_MOD_CTX, &ctx)) return -errno; return 0; } @@ -1387,7 +1400,7 @@ int drmGetContextFlags(int fd, drm_context_t context, drm_ctx_t ctx; ctx.handle = context; - if (ioctl(fd, DRM_IOCTL_GET_CTX, &ctx)) + if (drmIoctl(fd, DRM_IOCTL_GET_CTX, &ctx)) return -errno; *flags = 0; if (ctx.flags & _DRM_CONTEXT_PRESERVED) @@ -1418,7 +1431,7 @@ int drmDestroyContext(int fd, drm_context_t handle) { drm_ctx_t ctx; ctx.handle = handle; - if (ioctl(fd, DRM_IOCTL_RM_CTX, &ctx)) + if (drmIoctl(fd, DRM_IOCTL_RM_CTX, &ctx)) return -errno; return 0; } @@ -1426,7 +1439,7 @@ int drmDestroyContext(int fd, drm_context_t handle) int drmCreateDrawable(int fd, drm_drawable_t *handle) { drm_draw_t draw; - if (ioctl(fd, DRM_IOCTL_ADD_DRAW, &draw)) + if (drmIoctl(fd, DRM_IOCTL_ADD_DRAW, &draw)) return -errno; *handle = draw.handle; return 0; @@ -1436,7 +1449,7 @@ int drmDestroyDrawable(int fd, drm_drawable_t handle) { drm_draw_t draw; draw.handle = handle; - if (ioctl(fd, DRM_IOCTL_RM_DRAW, &draw)) + if (drmIoctl(fd, DRM_IOCTL_RM_DRAW, &draw)) return -errno; return 0; } @@ -1452,7 +1465,7 @@ int drmUpdateDrawableInfo(int fd, drm_drawable_t handle, update.num = num; update.data = (unsigned long long)(unsigned long)data; - if (ioctl(fd, DRM_IOCTL_UPDATE_DRAW, &update)) + if (drmIoctl(fd, DRM_IOCTL_UPDATE_DRAW, &update)) return -errno; return 0; @@ -1472,7 +1485,7 @@ int drmUpdateDrawableInfo(int fd, drm_drawable_t handle, */ int drmAgpAcquire(int fd) { - if (ioctl(fd, DRM_IOCTL_AGP_ACQUIRE, NULL)) + if (drmIoctl(fd, DRM_IOCTL_AGP_ACQUIRE, NULL)) return -errno; return 0; } @@ -1490,7 +1503,7 @@ int drmAgpAcquire(int fd) */ int drmAgpRelease(int fd) { - if (ioctl(fd, DRM_IOCTL_AGP_RELEASE, NULL)) + if (drmIoctl(fd, DRM_IOCTL_AGP_RELEASE, NULL)) return -errno; return 0; } @@ -1513,7 +1526,7 @@ int drmAgpEnable(int fd, unsigned long mode) drm_agp_mode_t m; m.mode = mode; - if (ioctl(fd, DRM_IOCTL_AGP_ENABLE, &m)) + if (drmIoctl(fd, DRM_IOCTL_AGP_ENABLE, &m)) return -errno; return 0; } @@ -1544,7 +1557,7 @@ int drmAgpAlloc(int fd, unsigned long size, unsigned long type, b.size = size; b.handle = 0; b.type = type; - if (ioctl(fd, DRM_IOCTL_AGP_ALLOC, &b)) + if (drmIoctl(fd, DRM_IOCTL_AGP_ALLOC, &b)) return -errno; if (address != 0UL) *address = b.physical; @@ -1571,7 +1584,7 @@ int drmAgpFree(int fd, drm_handle_t handle) b.size = 0; b.handle = handle; - if (ioctl(fd, DRM_IOCTL_AGP_FREE, &b)) + if (drmIoctl(fd, DRM_IOCTL_AGP_FREE, &b)) return -errno; return 0; } @@ -1596,7 +1609,7 @@ int drmAgpBind(int fd, drm_handle_t handle, unsigned long offset) b.handle = handle; b.offset = offset; - if (ioctl(fd, DRM_IOCTL_AGP_BIND, &b)) + if (drmIoctl(fd, DRM_IOCTL_AGP_BIND, &b)) return -errno; return 0; } @@ -1620,7 +1633,7 @@ int drmAgpUnbind(int fd, drm_handle_t handle) b.handle = handle; b.offset = 0; - if (ioctl(fd, DRM_IOCTL_AGP_UNBIND, &b)) + if (drmIoctl(fd, DRM_IOCTL_AGP_UNBIND, &b)) return -errno; return 0; } @@ -1641,7 +1654,7 @@ int drmAgpVersionMajor(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return -errno; return i.agp_version_major; } @@ -1662,7 +1675,7 @@ int drmAgpVersionMinor(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return -errno; return i.agp_version_minor; } @@ -1683,7 +1696,7 @@ unsigned long drmAgpGetMode(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.mode; } @@ -1704,7 +1717,7 @@ unsigned long drmAgpBase(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.aperture_base; } @@ -1725,7 +1738,7 @@ unsigned long drmAgpSize(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.aperture_size; } @@ -1746,7 +1759,7 @@ unsigned long drmAgpMemoryUsed(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.memory_used; } @@ -1767,7 +1780,7 @@ unsigned long drmAgpMemoryAvail(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.memory_allowed; } @@ -1788,7 +1801,7 @@ unsigned int drmAgpVendorId(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.id_vendor; } @@ -1809,7 +1822,7 @@ unsigned int drmAgpDeviceId(int fd) { drm_agp_info_t i; - if (ioctl(fd, DRM_IOCTL_AGP_INFO, &i)) + if (drmIoctl(fd, DRM_IOCTL_AGP_INFO, &i)) return 0; return i.id_device; } @@ -1821,7 +1834,7 @@ int drmScatterGatherAlloc(int fd, unsigned long size, drm_handle_t *handle) *handle = 0; sg.size = size; sg.handle = 0; - if (ioctl(fd, DRM_IOCTL_SG_ALLOC, &sg)) + if (drmIoctl(fd, DRM_IOCTL_SG_ALLOC, &sg)) return -errno; *handle = sg.handle; return 0; @@ -1833,7 +1846,7 @@ int drmScatterGatherFree(int fd, drm_handle_t handle) sg.size = 0; sg.handle = handle; - if (ioctl(fd, DRM_IOCTL_SG_FREE, &sg)) + if (drmIoctl(fd, DRM_IOCTL_SG_FREE, &sg)) return -errno; return 0; } @@ -1854,7 +1867,7 @@ int drmWaitVBlank(int fd, drmVBlankPtr vbl) int ret; do { - ret = ioctl(fd, DRM_IOCTL_WAIT_VBLANK, vbl); + ret = drmIoctl(fd, DRM_IOCTL_WAIT_VBLANK, vbl); vbl->request.type &= ~DRM_VBLANK_RELATIVE; } while (ret && errno == EINTR); @@ -1904,7 +1917,7 @@ int drmCtlInstHandler(int fd, int irq) ctl.func = DRM_INST_HANDLER; ctl.irq = irq; - if (ioctl(fd, DRM_IOCTL_CONTROL, &ctl)) + if (drmIoctl(fd, DRM_IOCTL_CONTROL, &ctl)) return -errno; return 0; } @@ -1927,7 +1940,7 @@ int drmCtlUninstHandler(int fd) ctl.func = DRM_UNINST_HANDLER; ctl.irq = 0; - if (ioctl(fd, DRM_IOCTL_CONTROL, &ctl)) + if (drmIoctl(fd, DRM_IOCTL_CONTROL, &ctl)) return -errno; return 0; } @@ -1944,7 +1957,7 @@ int drmFinish(int fd, int context, drmLockFlags flags) if (flags & DRM_LOCK_FLUSH_ALL) lock.flags |= _DRM_LOCK_FLUSH_ALL; if (flags & DRM_HALT_ALL_QUEUES) lock.flags |= _DRM_HALT_ALL_QUEUES; if (flags & DRM_HALT_CUR_QUEUES) lock.flags |= _DRM_HALT_CUR_QUEUES; - if (ioctl(fd, DRM_IOCTL_FINISH, &lock)) + if (drmIoctl(fd, DRM_IOCTL_FINISH, &lock)) return -errno; return 0; } @@ -1970,7 +1983,7 @@ int drmGetInterruptFromBusID(int fd, int busnum, int devnum, int funcnum) p.busnum = busnum; p.devnum = devnum; p.funcnum = funcnum; - if (ioctl(fd, DRM_IOCTL_IRQ_BUSID, &p)) + if (drmIoctl(fd, DRM_IOCTL_IRQ_BUSID, &p)) return -errno; return p.irq; } @@ -2012,7 +2025,7 @@ int drmAddContextPrivateMapping(int fd, drm_context_t ctx_id, map.ctx_id = ctx_id; map.handle = (void *)handle; - if (ioctl(fd, DRM_IOCTL_SET_SAREA_CTX, &map)) + if (drmIoctl(fd, DRM_IOCTL_SET_SAREA_CTX, &map)) return -errno; return 0; } @@ -2024,7 +2037,7 @@ int drmGetContextPrivateMapping(int fd, drm_context_t ctx_id, map.ctx_id = ctx_id; - if (ioctl(fd, DRM_IOCTL_GET_SAREA_CTX, &map)) + if (drmIoctl(fd, DRM_IOCTL_GET_SAREA_CTX, &map)) return -errno; if (handle) *handle = (drm_handle_t)map.handle; @@ -2039,7 +2052,7 @@ int drmGetMap(int fd, int idx, drm_handle_t *offset, drmSize *size, drm_map_t map; map.offset = idx; - if (ioctl(fd, DRM_IOCTL_GET_MAP, &map)) + if (drmIoctl(fd, DRM_IOCTL_GET_MAP, &map)) return -errno; *offset = map.offset; *size = map.size; @@ -2056,7 +2069,7 @@ int drmGetClient(int fd, int idx, int *auth, int *pid, int *uid, drm_client_t client; client.idx = idx; - if (ioctl(fd, DRM_IOCTL_GET_CLIENT, &client)) + if (drmIoctl(fd, DRM_IOCTL_GET_CLIENT, &client)) return -errno; *auth = client.auth; *pid = client.pid; @@ -2071,7 +2084,7 @@ int drmGetStats(int fd, drmStatsT *stats) drm_stats_t s; int i; - if (ioctl(fd, DRM_IOCTL_GET_STATS, &s)) + if (drmIoctl(fd, DRM_IOCTL_GET_STATS, &s)) return -errno; stats->count = 0; @@ -2213,7 +2226,7 @@ int drmSetInterfaceVersion(int fd, drmSetVersion *version) sv.drm_dd_major = version->drm_dd_major; sv.drm_dd_minor = version->drm_dd_minor; - if (ioctl(fd, DRM_IOCTL_SET_VERSION, &sv)) { + if (drmIoctl(fd, DRM_IOCTL_SET_VERSION, &sv)) { retcode = -errno; } @@ -2244,7 +2257,7 @@ int drmCommandNone(int fd, unsigned long drmCommandIndex) request = DRM_IO( DRM_COMMAND_BASE + drmCommandIndex); - if (ioctl(fd, request, data)) { + if (drmIoctl(fd, request, data)) { return -errno; } return 0; @@ -2273,7 +2286,7 @@ int drmCommandRead(int fd, unsigned long drmCommandIndex, void *data, request = DRM_IOC( DRM_IOC_READ, DRM_IOCTL_BASE, DRM_COMMAND_BASE + drmCommandIndex, size); - if (ioctl(fd, request, data)) { + if (drmIoctl(fd, request, data)) { return -errno; } return 0; @@ -2302,7 +2315,7 @@ int drmCommandWrite(int fd, unsigned long drmCommandIndex, void *data, request = DRM_IOC( DRM_IOC_WRITE, DRM_IOCTL_BASE, DRM_COMMAND_BASE + drmCommandIndex, size); - if (ioctl(fd, request, data)) { + if (drmIoctl(fd, request, data)) { return -errno; } return 0; @@ -2331,9 +2344,8 @@ int drmCommandWriteRead(int fd, unsigned long drmCommandIndex, void *data, request = DRM_IOC( DRM_IOC_READ|DRM_IOC_WRITE, DRM_IOCTL_BASE, DRM_COMMAND_BASE + drmCommandIndex, size); - if (ioctl(fd, request, data)) { + if (drmIoctl(fd, request, data)) return -errno; - } return 0; } @@ -2355,7 +2367,7 @@ int drmFenceCreate(int fd, unsigned flags, int fence_class, unsigned type, arg.type = type; arg.fence_class = fence_class; - if (ioctl(fd, DRM_IOCTL_FENCE_CREATE, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_CREATE, &arg)) return -errno; fence->handle = arg.handle; fence->fence_class = arg.fence_class; @@ -2379,7 +2391,7 @@ int drmFenceBuffers(int fd, unsigned flags, uint32_t fence_class, drmFence *fenc arg.flags = flags; arg.fence_class = fence_class; - if (ioctl(fd, DRM_IOCTL_FENCE_BUFFERS, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_BUFFERS, &arg)) return -errno; fence->handle = arg.handle; fence->fence_class = arg.fence_class; @@ -2397,7 +2409,7 @@ int drmFenceReference(int fd, unsigned handle, drmFence *fence) memset(&arg, 0, sizeof(arg)); arg.handle = handle; - if (ioctl(fd, DRM_IOCTL_FENCE_REFERENCE, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_REFERENCE, &arg)) return -errno; fence->handle = arg.handle; fence->fence_class = arg.fence_class; @@ -2414,7 +2426,7 @@ int drmFenceUnreference(int fd, const drmFence *fence) memset(&arg, 0, sizeof(arg)); arg.handle = fence->handle; - if (ioctl(fd, DRM_IOCTL_FENCE_UNREFERENCE, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_UNREFERENCE, &arg)) return -errno; return 0; } @@ -2427,7 +2439,7 @@ int drmFenceFlush(int fd, drmFence *fence, unsigned flush_type) arg.handle = fence->handle; arg.type = flush_type; - if (ioctl(fd, DRM_IOCTL_FENCE_FLUSH, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_FLUSH, &arg)) return -errno; fence->fence_class = arg.fence_class; fence->type = arg.type; @@ -2442,7 +2454,7 @@ int drmFenceUpdate(int fd, drmFence *fence) memset(&arg, 0, sizeof(arg)); arg.handle = fence->handle; - if (ioctl(fd, DRM_IOCTL_FENCE_SIGNALED, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_SIGNALED, &arg)) return -errno; fence->fence_class = arg.fence_class; fence->type = arg.type; @@ -2482,7 +2494,7 @@ int drmFenceEmit(int fd, unsigned flags, drmFence *fence, unsigned emit_type) arg.handle = fence->handle; arg.type = emit_type; - if (ioctl(fd, DRM_IOCTL_FENCE_EMIT, &arg)) + if (drmIoctl(fd, DRM_IOCTL_FENCE_EMIT, &arg)) return -errno; fence->fence_class = arg.fence_class; fence->type = arg.type; @@ -2520,7 +2532,7 @@ drmIoctlTimeout(int fd, unsigned long request, void *argp) int ret; do { - ret = ioctl(fd, request, argp); + ret = drmIoctl(fd, request, argp); if (ret != 0 && errno == EAGAIN) { if (!haveThen) { gettimeofday(&then, NULL); @@ -2630,7 +2642,7 @@ int drmBOReference(int fd, unsigned handle, drmBO *buf) memset(&arg, 0, sizeof(arg)); req->handle = handle; - if (ioctl(fd, DRM_IOCTL_BO_REFERENCE, &arg)) + if (drmIoctl(fd, DRM_IOCTL_BO_REFERENCE, &arg)) return -errno; drmBOCopyReply(rep, buf); @@ -2654,7 +2666,7 @@ int drmBOUnreference(int fd, drmBO *buf) memset(&arg, 0, sizeof(arg)); arg.handle = buf->handle; - if (ioctl(fd, DRM_IOCTL_BO_UNREFERENCE, &arg)) + if (drmIoctl(fd, DRM_IOCTL_BO_UNREFERENCE, &arg)) return -errno; buf->handle = 0; @@ -2724,7 +2736,7 @@ int drmBOUnmap(int fd, drmBO *buf) memset(&arg, 0, sizeof(arg)); arg.handle = buf->handle; - if (ioctl(fd, DRM_IOCTL_BO_UNMAP, &arg)) { + if (drmIoctl(fd, DRM_IOCTL_BO_UNMAP, &arg)) { return -errno; } buf->mapCount--; @@ -2770,7 +2782,7 @@ int drmBOInfo(int fd, drmBO *buf) memset(&arg, 0, sizeof(arg)); req->handle = buf->handle; - ret = ioctl(fd, DRM_IOCTL_BO_INFO, &arg); + ret = drmIoctl(fd, DRM_IOCTL_BO_INFO, &arg); if (ret) return -errno; @@ -2825,7 +2837,7 @@ int drmMMInit(int fd, unsigned long pOffset, unsigned long pSize, arg.p_size = pSize; arg.mem_type = memType; - if (ioctl(fd, DRM_IOCTL_MM_INIT, &arg)) + if (drmIoctl(fd, DRM_IOCTL_MM_INIT, &arg)) return -errno; return 0; } @@ -2837,7 +2849,7 @@ int drmMMTakedown(int fd, unsigned memType) memset(&arg, 0, sizeof(arg)); arg.mem_type = memType; - if (ioctl(fd, DRM_IOCTL_MM_TAKEDOWN, &arg)) + if (drmIoctl(fd, DRM_IOCTL_MM_TAKEDOWN, &arg)) return -errno; return 0; } @@ -2879,7 +2891,7 @@ int drmMMInfo(int fd, unsigned memType, uint64_t *size) arg.mem_type = memType; - if (ioctl(fd, DRM_IOCTL_MM_INFO, &arg)) + if (drmIoctl(fd, DRM_IOCTL_MM_INFO, &arg)) return -errno; *size = arg.p_size; @@ -2894,7 +2906,7 @@ int drmBOVersion(int fd, unsigned int *major, int ret; memset(&arg, 0, sizeof(arg)); - ret = ioctl(fd, DRM_IOCTL_BO_VERSION, &arg); + ret = drmIoctl(fd, DRM_IOCTL_BO_VERSION, &arg); if (ret) return -errno; -- cgit v1.2.3 From 73bc18cad8d1c6b4481a199cebf7f0a28d19c2bb Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 17:06:35 -0700 Subject: [intel-gem] Wait for rendering to complete before unbinding. Moving to the CPU domain doesn't ensure that rendering is finished, the buffer may still be in use as a texture or other data source. --- linux-core/i915_gem.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index cb5d663e..0f037acf 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -688,6 +688,19 @@ i915_gem_object_unbind(struct drm_gem_object *obj) 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 -- cgit v1.2.3 From a7139cb8511a9d31d9f79bcaae62020d30e09f90 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 19:35:22 -0700 Subject: [intel-gem] show total GTT space in /proc/dri/*/gem_objects --- linux-core/drmP.h | 1 + linux-core/drm_proc.c | 11 ++++++----- linux-core/i915_gem.c | 2 ++ 3 files changed, 9 insertions(+), 5 deletions(-) diff --git a/linux-core/drmP.h b/linux-core/drmP.h index ba25f3a8..392d2ace 100644 --- a/linux-core/drmP.h +++ b/linux-core/drmP.h @@ -970,6 +970,7 @@ struct drm_device { atomic_t pin_memory; atomic_t gtt_count; atomic_t gtt_memory; + uint32_t gtt_total; uint32_t invalidate_domains; /* domains pending invalidation */ uint32_t flush_domains; /* domains pending flush */ /*@} */ diff --git a/linux-core/drm_proc.c b/linux-core/drm_proc.c index 71d55274..2bbe7eea 100644 --- a/linux-core/drm_proc.c +++ b/linux-core/drm_proc.c @@ -654,11 +654,12 @@ static int drm_gem_object_info(char *buf, char **start, off_t offset, *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 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; diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 0f037acf..e07cf1b9 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -65,6 +65,8 @@ i915_gem_init_ioctl(struct drm_device *dev, void *data, drm_memrange_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; -- cgit v1.2.3 From 732b1960742042eb33f49c2b3cdd2d36eadbc920 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 19:37:44 -0700 Subject: [intel-gem] whitespace fixes --- linux-core/i915_gem.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index e07cf1b9..78d0d952 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -66,7 +66,7 @@ i915_gem_init_ioctl(struct drm_device *dev, void *data, args->gtt_end - args->gtt_start); dev->gtt_total = (uint32_t) (args->gtt_end - args->gtt_start); - + mutex_unlock(&dev->struct_mutex); return 0; @@ -500,7 +500,7 @@ i915_gem_retire_requests(struct drm_device *dev) list_del(&request->list); drm_free(request, sizeof(*request), DRM_MEM_DRIVER); } else - break; + break; } } @@ -858,7 +858,7 @@ i915_gem_evict_something(struct drm_device *dev) list); ret = i915_wait_request(dev, request->seqno); - + /* if waiting caused an object to become inactive, * then loop around and wait for it. Otherwise, we * assume that waiting freed and unbound something, -- cgit v1.2.3 From 68856b619bc1a2e91e67764911c8af4e2466fad9 Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 19:40:16 -0700 Subject: [intel-gem] Debugging -- verify inactive list invariants Inactive list elements may not be pinned, active or have non-CPU write domains. --- linux-core/i915_gem.c | 40 +++++++++++++++++++++++++++++++++++++++- 1 file changed, 39 insertions(+), 1 deletion(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 78d0d952..8741eecd 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -35,6 +35,7 @@ #define WATCH_EXEC 0 #define WATCH_LRU 0 #define WATCH_RELOC 0 +#define WATCH_INACTIVE 0 static int i915_gem_object_set_domain(struct drm_gem_object *obj, @@ -303,6 +304,26 @@ i915_gem_object_move_to_active(struct drm_gem_object *obj) &dev_priv->mm.active_list); } +#if WATCH_INACTIVE +static 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)) + 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); + } +} +#else +#define i915_verify_inactive(dev,file,line) +#endif + static void i915_gem_object_move_to_inactive(struct drm_gem_object *obj) { @@ -310,6 +331,7 @@ i915_gem_object_move_to_inactive(struct drm_gem_object *obj) 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 @@ -319,6 +341,7 @@ i915_gem_object_move_to_inactive(struct drm_gem_object *obj) obj_priv->active = 0; drm_gem_object_unreference(obj); } + i915_verify_inactive(dev, __FILE__, __LINE__); } /** @@ -1635,6 +1658,7 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, mutex_lock(&dev->struct_mutex); + i915_verify_inactive(dev, __FILE__, __LINE__); if (dev_priv->mm.suspended) { DRM_ERROR("Execbuf while VT-switched.\n"); mutex_unlock(&dev->struct_mutex); @@ -1676,6 +1700,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, 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; @@ -1698,9 +1724,13 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, 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], @@ -1730,6 +1760,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, */ 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, @@ -1754,6 +1786,8 @@ i915_gem_execbuffer(struct drm_device *dev, void *data, 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, @@ -1789,6 +1823,7 @@ i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment) 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) { @@ -1807,6 +1842,7 @@ i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment) if (!obj_priv->active && obj->write_domain == 0) list_del_init(&obj_priv->list); } + i915_verify_inactive(dev, __FILE__, __LINE__); return 0; } @@ -1818,10 +1854,11 @@ i915_gem_object_unpin(struct drm_gem_object *obj) 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 @@ -1832,6 +1869,7 @@ i915_gem_object_unpin(struct drm_gem_object *obj) atomic_dec(&dev->pin_count); atomic_sub(obj->size, &dev->pin_memory); } + i915_verify_inactive(dev, __FILE__, __LINE__); } int -- cgit v1.2.3 From 93c2871eccc1abde0d88ea439cf963c4895a26fc Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 19:43:40 -0700 Subject: [intel-gem] BUG_ON active objects in gem_object_unbind Now that gem_object_unbind waits for rendering to complete, objects should not be active when they are being pulled from the GTT. BUG_ON if this is broken. --- linux-core/i915_gem.c | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 8741eecd..a15d7f23 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -734,8 +734,10 @@ i915_gem_object_unbind(struct drm_gem_object *obj) */ ret = i915_gem_object_set_domain(obj, I915_GEM_DOMAIN_CPU, I915_GEM_DOMAIN_CPU); - if (ret) + 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); @@ -743,6 +745,8 @@ i915_gem_object_unbind(struct drm_gem_object *obj) obj_priv->agp_mem = NULL; } + BUG_ON(obj_priv->active); + i915_gem_object_free_page_list(obj); atomic_dec(&dev->gtt_count); @@ -752,15 +756,9 @@ i915_gem_object_unbind(struct drm_gem_object *obj) obj_priv->gtt_space = NULL; /* Remove ourselves from the LRU list if present. */ - if (!list_empty(&obj_priv->list)) { + if (!list_empty(&obj_priv->list)) list_del_init(&obj_priv->list); - if (obj_priv->active) { - DRM_ERROR("Failed to wait on buffer when unbinding, " - "continued anyway.\n"); - obj_priv->active = 0; - drm_gem_object_unreference(obj); - } - } + return 0; } -- cgit v1.2.3 From 19c3418848ccdbb163cd16b354b14b0559813d6c Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 19:47:23 -0700 Subject: [intel-gem] inactive list may contain objects in CPU write domain Pin/unpin need to know whether to remove/add objects from the inactive list, inactive objects cannot be in any GPU write domain as those would be on the flushing list instead. However, inactive objects may be in the CPU write domain. --- linux-core/i915_gem.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index a15d7f23..5e5c0ea1 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -1837,7 +1837,8 @@ i915_gem_object_pin(struct drm_gem_object *obj, uint32_t alignment) 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 == 0) + if (!obj_priv->active && (obj->write_domain & ~I915_GEM_DOMAIN_CPU) == 0 && + !list_empty(&obj_priv->list)) list_del_init(&obj_priv->list); } i915_verify_inactive(dev, __FILE__, __LINE__); @@ -1862,8 +1863,9 @@ i915_gem_object_unpin(struct drm_gem_object *obj) * the inactive list */ if (obj_priv->pin_count == 0) { - if (!obj_priv->active && obj->write_domain == 0) - list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list); + if (!obj_priv->active && (obj->write_domain & ~I915_GEM_DOMAIN_CPU) == 0) + list_move_tail(&obj_priv->list, + &dev_priv->mm.inactive_list); atomic_dec(&dev->pin_count); atomic_sub(obj->size, &dev->pin_memory); } -- cgit v1.2.3 From 3e48e144992fb11b31875989d45bc8a7c041cdef Mon Sep 17 00:00:00 2001 From: Keith Packard Date: Fri, 13 Jun 2008 19:49:47 -0700 Subject: [intel-gem] Execute MI_FLUSH in leavevt_ioctl In leavevt_ioctl, queue an MI_FLUSH and then block waiting for it to complete. This will empty the active and flushing lists. That leaves only the inactive list to evict. --- linux-core/i915_gem.c | 52 ++++++++++++++++++++++++++++++++------------------- 1 file changed, 33 insertions(+), 19 deletions(-) diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c index 5e5c0ea1..308674d8 100644 --- a/linux-core/i915_gem.c +++ b/linux-core/i915_gem.c @@ -2127,6 +2127,10 @@ i915_gem_entervt_ioctl(struct drm_device *dev, void *data, 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; @@ -2170,6 +2174,8 @@ i915_gem_leavevt_ioctl(struct drm_device *dev, void *data, struct drm_file *file_priv) { drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t seqno; + int ret; mutex_lock(&dev->struct_mutex); /* Hack! Don't let anybody do execbuf while we don't control the chip. @@ -2177,32 +2183,40 @@ i915_gem_leavevt_ioctl(struct drm_device *dev, void *data, */ dev_priv->mm.suspended = 1; - /* Move all buffers out of the GTT. */ - i915_gem_evict_from_list(dev, &dev_priv->mm.active_list); - i915_gem_evict_from_list(dev, &dev_priv->mm.flushing_list); - i915_gem_evict_from_list(dev, &dev_priv->mm.inactive_list); + i915_kernel_lost_context(dev); - /* Make sure the harware's idle. */ - while (!list_empty(&dev_priv->mm.request_list)) { - struct drm_i915_gem_request *request; - int ret; + /* Flush the GPU along with all non-CPU write domains + */ + i915_gem_flush(dev, ~I915_GEM_DOMAIN_CPU, ~I915_GEM_DOMAIN_CPU); + seqno = i915_add_request(dev, ~I915_GEM_DOMAIN_CPU); + if (seqno == 0) { + mutex_unlock(&dev->struct_mutex); + return -ENOMEM; + } + ret = i915_wait_request(dev, seqno); + if (ret) { + mutex_unlock(&dev->struct_mutex); + return ret; + } - request = list_first_entry(&dev_priv->mm.request_list, - struct drm_i915_gem_request, - list); + /* 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)); - ret = i915_wait_request(dev, request->seqno); - if (ret != 0) { - DRM_ERROR("Error waiting for idle at LeaveVT: %d\n", - ret); - mutex_unlock(&dev->struct_mutex); - return ret; - } - } + /* 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. */ + i915_gem_evict_from_list(dev, &dev_priv->mm.inactive_list); 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)); i915_gem_cleanup_ringbuffer(dev); -- cgit v1.2.3 From c847271179da382af52a6874e2edec586b88af75 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Mon, 16 Jun 2008 13:09:31 -0700 Subject: [linux] Use the device's irq for handler setup instead of stale dev->irq. This fixes registration when MSI is set up after the stub function fills in dev->irq. Otherwise /proc/interrupts would report attachment to the fasteoi interrupt. dev->irq is still exposed (and updated at IRQ setup) for the drivers that use it for whatever reason. --- linux-core/drm_irq.c | 20 ++++++++++++-------- shared-core/i915_dma.c | 4 ++-- 2 files changed, 14 insertions(+), 10 deletions(-) diff --git a/linux-core/drm_irq.c b/linux-core/drm_irq.c index 318d9d7a..961f90c4 100644 --- a/linux-core/drm_irq.c +++ b/linux-core/drm_irq.c @@ -63,7 +63,7 @@ int drm_irq_by_busid(struct drm_device *dev, void *data, p->devnum != PCI_SLOT(dev->pdev->devfn) || p->funcnum != PCI_FUNC(dev->pdev->devfn)) return -EINVAL; - p->irq = dev->irq; + p->irq = dev->pdev->irq; DRM_DEBUG("%d:%d:%d => IRQ %d\n", p->busnum, p->devnum, p->funcnum, p->irq); @@ -200,7 +200,7 @@ int drm_irq_install(struct drm_device * dev) if (!drm_core_check_feature(dev, DRIVER_HAVE_IRQ)) return -EINVAL; - if (dev->irq == 0) + if (dev->pdev->irq == 0) return -EINVAL; mutex_lock(&dev->struct_mutex); @@ -218,7 +218,7 @@ int drm_irq_install(struct drm_device * dev) dev->irq_enabled = 1; mutex_unlock(&dev->struct_mutex); - DRM_DEBUG("irq=%d\n", dev->irq); + DRM_DEBUG("irq=%d\n", dev->pdev->irq); /* Before installing handler */ dev->driver->irq_preinstall(dev); @@ -227,7 +227,7 @@ int drm_irq_install(struct drm_device * dev) if (drm_core_check_feature(dev, DRIVER_IRQ_SHARED)) sh_flags = IRQF_SHARED; - ret = request_irq(dev->irq, dev->driver->irq_handler, + ret = request_irq(dev->pdev->irq, dev->driver->irq_handler, sh_flags, dev->devname, dev); if (ret < 0) { mutex_lock(&dev->struct_mutex); @@ -235,6 +235,10 @@ int drm_irq_install(struct drm_device * dev) mutex_unlock(&dev->struct_mutex); return ret; } + /* Expose the device irq to device drivers that want to export it for + * whatever reason. + */ + dev->irq = dev->pdev->irq; /* After installing handler */ ret = dev->driver->irq_postinstall(dev); @@ -270,11 +274,11 @@ int drm_irq_uninstall(struct drm_device * dev) if (!irq_enabled) return -EINVAL; - DRM_DEBUG("irq=%d\n", dev->irq); + DRM_DEBUG("irq=%d\n", dev->pdev->irq); dev->driver->irq_uninstall(dev); - free_irq(dev->irq, dev); + free_irq(dev->pdev->irq, dev); drm_vblank_cleanup(dev); @@ -308,7 +312,7 @@ int drm_control(struct drm_device *dev, void *data, if (!drm_core_check_feature(dev, DRIVER_HAVE_IRQ)) return 0; if (dev->if_version < DRM_IF_VERSION(1, 2) && - ctl->irq != dev->irq) + ctl->irq != dev->pdev->irq) return -EINVAL; return drm_irq_install(dev); case DRM_UNINST_HANDLER: @@ -489,7 +493,7 @@ int drm_wait_vblank(struct drm_device *dev, void *data, int ret = 0; unsigned int flags, seq, crtc; - if ((!dev->irq) || (!dev->irq_enabled)) + if ((!dev->pdev->irq) || (!dev->irq_enabled)) return -EINVAL; if (vblwait->request.type & diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c index 2157c594..2db6be07 100644 --- a/shared-core/i915_dma.c +++ b/shared-core/i915_dma.c @@ -150,7 +150,7 @@ static int i915_dma_cleanup(struct drm_device * dev) * may not have been called from userspace and after dev_private * is freed, it's too late. */ - if (dev->irq) + if (dev->irq_enabled) drm_irq_uninstall(dev); if (dev_priv->ring.virtual_start) { @@ -889,7 +889,7 @@ static int i915_getparam(struct drm_device *dev, void *data, switch (param->param) { case I915_PARAM_IRQ_ACTIVE: - value = dev->irq ? 1 : 0; + value = dev->irq_enabled ? 1 : 0; break; case I915_PARAM_ALLOW_BATCHBUFFER: value = dev_priv->allow_batchbuffer ? 1 : 0; -- cgit v1.2.3 From e7424e4580159b0ac3e232674dff5c862e851dff Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Mon, 16 Jun 2008 15:15:02 -0700 Subject: [intel] Quirk away MSI support on 945G/GM. The PCI caps register reports MSI support even though it isn't really there. --- linux-core/i915_drv.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/linux-core/i915_drv.c b/linux-core/i915_drv.c index 012ca822..e399f374 100644 --- a/linux-core/i915_drv.c +++ b/linux-core/i915_drv.c @@ -628,7 +628,16 @@ static int probe(struct pci_dev *pdev, const struct pci_device_id *ent) { int ret; - (void) pci_enable_msi(pdev); + /* On the 945G/GM, the chipset reports the MSI capability on the + * integrated graphics even though the support isn't actually there + * according to the published specs. It doesn't appear to function + * correctly in testing on 945G. + * This may be a side effect of MSI having been made available for PEG + * and the registers being closely associated. + */ + if (pdev->device != 0x2772 && pdev->device != 0x27A2) + (void )pci_enable_msi(pdev); + ret = drm_get_dev(pdev, ent, &driver); if (ret && pdev->msi_enabled) pci_disable_msi(pdev); -- cgit v1.2.3